(1) 信息识别检测方面的背景
(2) 红外识别(IRID)技术的背景(特点\基本原理\所需技术\面部温谱图)
(3) (面部特征变量)实验变量(面部温谱图è环境温度\气流状况\代谢活动\面表温度\局部因素等)è变量的舍取è建立分级数据库
(4) 热图形的特征提取
(过去使用的方法及其特点è度规比配(Metrics Matching)\ 面部模板匹配(眼角区域和脸颊内侧)\ 基本形状实时匹配\基本轮廓的后期匹配\小波分析技术(系统组成\))
文章介绍了各种方法中数据库的大小、图像种类及来源、测试结果等
(最近的研究现状及发展方向：使用制冷或非制冷的红外摄像机，基于解剖学结构特征
[1] 对称波形技术è（介绍了六种技术）首先对人脸进行处理，在假设脸部近似对称的情况下，在脸部选取多个特征点，赋值并构造波形图，并从中提取数据。仅适用于小型或分级数据库；
[2] 面部编码技术 从对称波形图中提取一组一维条形码。另外还介绍了一种二维编码技术和将来更先进的编码技术（红外脸部图像编码技术）。
（热细节特征点及可视细节特征点è介绍了一种基于红外和可见光人脸图像的相互联系IRID系统及其应用领域，重点介绍了在视频压缩传输、卫生医疗等方面的应用情况）
(5) 红外识别系统的应用领域（五个：进出口管制、计算机安全、智能监控、Talk-Head 视频压缩、卫生保健）
(6) 红外识别技术未来的应用领域（随身的IRID系统、通信、可增强语音识别功能、情感识别等）
Notes
《History,current status,and feature if infrared identification》（红外鉴定的历史、现状及特征）1.Abstract
Infrared identification (IRID) is a biometric which offers the security of fingerprints with the convenience of facial recognition（人脸识别）. Twenty years after being first proposed, IRID systems are becoming marketable. During the intervening time, IR camera technology（红外摄像机技术）has improved significantly and price reductions have been dramatic. Further improvements in IR camera sensitivity, array size, and variable optics, as well as further price reductions, are coming, just as interest is growing in biometrics. Several approaches have been developed, tested, and evaluated to extract and analyze（识别与分析） features from infrared images of the face and body. IRID has been to have significant advantages over other biometrics in certain applications --- not only in the traditional biometric security markets, but also in communications and health care（卫生保健）.
2. Introduction
The general public has accepted the need for increased security measures. In fact they expect to see surveillance cameras (监视摄像机)in public areas, have learned to accept long waits for security checks at airports, and are accustomed to being asked for ID in order to cash checks, make withdrawals, obtain medical benefits, and pick up their children from day care. People are increasingly willing to pay for security by spending dollars and time, and by sacrificing some privacy. The market for security technology is therefore experiencing major growth. Biometrics is the fastest growing segment of the security industry. Fingerprint readers, handwriting verification, facial recognition, hand geometry, retinal and iris scanners, DNA analysis, and voice recognition are all becoming familiar techniques for identifying who a person is, and for providing links into databases of personal information. For some applications, biometric identification is simply more convenient than remembering a password or carrying a keycard. However, for many applications accurate and rapid automatic identification is of serious importance.
(当今社会人们在某些场所为了安全起见，甘愿牺牲自己的某些隐私来获得一个相对安全的环境，因此安全技术市场也将面临着巨大的发展。生物技术是在安全领域中快速发展的一个分支。指纹识别、字迹核对、人脸识别、手掌几何特征、视网膜与虹膜扫描、DNA分析、声音识别都是身份验证及更新个人信息数据库的一些技术。在一些应用中，生物识别技术比起通过记忆某些验证密码并要携带验证物件要方便得多。然而在某些应用中，进行精确和快速自动验证是十分必要的。)
3. Background on IRID Technology
The goal of developing infrared identification has been to develop a technique which is as unique as fingerprints and as convenient as facial recognition. The anatomical information which is utilized by IRID involves subsurface features unique to each person. Those features may be imaged at a distance, using passive infrared sensor （被动（无源）红外传感器）technology, with or without the cooperation of the subject.. The thermal patterns seen in an IR image derive primarily from the pattern of superficial blood vessels which are those under the skin but above bone and muscle. The vessels transport warm blood throughout the body, and heat the skin above. Skin which is directly above a blood vessel is on the average
红外识别的特点：唯一性（指纹识别）+方便性（人脸识别）。红外识别所使用的人体表皮特征(subsurface features)的对每一个人来说是唯一的。这些特征信息可以通过在远处使用被动（无源）红外传感器技术，伴随或不伴随着对象的合作，捕获图像来获得。在红外图像中的热模式主要来源于位于皮肤下而骨骼与肌肉之上毛细血管的模式。这些血管将温热的血液传送与人体各部分，从而加热了上面的皮肤。直接位于血管上面的皮肤温度要比邻近的皮肤温度平均高
That range and sensitivity are well within the specifications of current IR detector technology. The diameter of a blood vessel is on the order of 0.1", which is well within the resolution of current focal plane arrays（焦平面阵列）and optics. The complexity and vastness of the layout of about 5 kilometers of blood vessels in the head and face assures that each person's vascular arrangement is irreproducible and hence unique. Even identical twins have different thermogram features. IRID provides the capability for rapid, on-the-fly positive identification, under all lighting conditions including total darkness.
这个范围和灵敏度恰好符合当前红外探测器技术所规定的要求。一个血管的直径大约是
Problems with visual facial recognition in dim lighting conditions are exacerbated when the subject has dark skin. The result is a difference in accuracy of performance as a function of skin color. A visual face recognition system which is optimized for identification of light-skinned persons could be prone to higher false alarms among dark- skinned persons; a situation which is unacceptable. Infrared identification sacrifices color recognition for lighting independence, with the result that it provides color-blind identification with the same accuracy for everyone. When colors of hair, eyes, skin, clothing, etc. is essential to match eyewitness descriptions, a color band can be added to the camera to provide that information. Positive automated identification, however, should be performed from IR images.
在昏暗光照环境下，当检验对象有着较黑皮肤时，基于可见光的人脸识别的问题将突出表现出来了。这将导致在同一肤色函数下表现出不同的准确性。而基于红外图像的人脸识别系统则是独立于可见光环境的。
Fingerprint ridges are formed early in gestation, when the thin skin responds to the underlying developing vascular structure. Therefore both fingerprint features and facial thermogram features have their origin in the detailed network of blood vessels and finer structures of the cardiovascular system. A comparison between minutiae from facial thermograms and from fingerprints based upon the anatomical structures underlying each supports the experimental findings that facial thermograms are unique, as fingerprints have long been assumed to be unique for each person. For many biometrics applications, identification based upon facial thermograms is preferable to use of fingerprints, since the former require no physical contact with the subject and can be collected on-the-fly.
指纹的凸起在婴儿孕育早期形成，此时形成的薄皮肤也已经根据潜在形成的血管结构做出了反应。因此，指纹特征和面部温谱图特征在详细的血管网络和出色的心血管系统结构方面是在初始阶段形成的。在细节特征点方面将提供者的指纹和面部温谱图解剖学结构上进行对比，试验结果表明，如同每个人的指纹是唯一的一样，人的温谱图也是唯一的。在生物识别应用方面，红外人脸识别技术要优于指纹识别，因为前者无需对检测者进行人体接触并且可以实现远程检测。
While there is no way to prove that facial thermograms are unique, it is possible to show they contain inherently more variations than fingerprints. Similarly, it has never been proven that fingerprints are unique. However, that working hypothesis has been supported by many years of experience. Facial thermogram minutiae is analogous to fingerprint minutiae in that two sets of minutiae may be considered to identify the same person if a significant number of the minutiae in the two sets have corresponding positions and characteristics.
尽管没有方法来证明人的面部温谱图是唯一的，但可以说明面部温谱图较指纹包含的变量要多。类似的，一直以来人们也无法证明每个人的指纹是唯一的。然而这一假说以尽为人们接受多年了。温谱图细节特征点类似于指纹细节特征点，因为对于给定的两组细节特征点特征如果存在大量的一致点的话就可以可以确定一个人的身份。
Classification of facial thermograms can be performed to partition a database and reduce the search for matching facial patterns, just as fingerprints may be classified as to type prior to searching a database. In the
面部温谱图的恰当分类可以减少搜寻数据库所花费的时间.在美国必须对16个细微点进行匹配才能完成一个指纹识别过程。在红外人脸识别方面简历相关的
4. Experimental Variables （实验变量）
Facial thermograms reflect variations in apparent temperatures across the facial surface which are produced by changes in ambient temperature and air flow conditions, as well as by exercise, metabolic activity associated with ingesting food and drink, illness, and drugs. Many other factors can produce localized changes.These include wearing tight clothing, being under stress,having a migraine headache, blushing, or having an infected tooth. The high level of variability in facial thermograms, representing the extreme level of sensitivity to a large number of variables, presents a challenge to find features which are persistent enough to use in automated identification.
面部问谱图反映了周围温度和气流状况的改变，运动和饮食引起的代谢活动，疾病以及药物等所引起的人脸表面的温度变化。还有一些因素会引起局部性变化，这些因素包括：穿较紧的衣服、处在一定的压力之下、患有偏头痛、脸红或者牙齿感染等。面部温谱图中变量的高度复杂性（表现为大量变量的极度敏感性）对发现足够稳定的人脸特征用于自动识别提出了挑战。
The solution is to ignore the apparent temperature values and use only the anatomical features they reflect which are invariant to changes in physiological and ambient conditions. Consider this analogous to analyzing fingerprints; whether the print is formed by coating the finger with oil, blood, blue ink, or chocolate, and whethera rolled or latent print is seen, the same fingerprint details emerge. Similarly, facial thermograms yield the same blood vessel pathways and minutiae regardless of apparent temperatures.
该问题的解决方法是忽略表面温度值而只使用他们反映的不随周围环境和生理特征改变的解剖学特征方面的变量。类似于分析人的指纹一样，不管该指纹是由涂有油脂、血液、蓝墨水还是巧克力的手指形成的，都会显示相同的指纹特征。同样忽略表面温度，面部温谱图记录了同样的血管路径和细节特征点特征。
Only a single frame of infrared video is required to uniquely identify a person if it is taken from the same aspect angle as a known reference image. IR images can be taken at the same time and with the same care as a standard photograph used for passports. There are strong correlations between the infrared and visual facial images:head shape and size, location and shape and size of features.The database of images can be segmented into classes using those values, and the same classification system will work for visual or infrared images
只需要红外录像中的一帧图像就可以唯一的确认一个人（该图像必须和参考图像从同一角度拍摄）。该红外图像可以在办理护照或其他证件时和标准图像同时拍摄。红外图像和可见光图像有很大的联系：头部形状和大小、面部分布、大小和形状等特征。图像数据库可以根据这些特征分为几类，根据分类确定使用可见光图像还是红外图像。
4. Feature Extraction from Thermal Images（热成像的特征提取）
IRID integrates infrared imaging, anatomical modeling, pattern recognition, and image compression to achieve a biometric capability. The initial research approaches utilized thermal contours for identification, and matched them through template, fractal, or wavelet methods. That approach is tailored for cooperative access control applications in which the subject voluntarily interacts with the system, such as at an ATM or point of sales (POS) terminal. Its accuracy is affected by activities which disturb the edges of the contours. This includes eyeglasses being put on or off or changed. It also includes variations in head position relative to the camera, and changes in facial and cranial hair.
红外图像识别结合了红外成像、结构建模、模式识别和图像压缩来实现生物认证功能。开始的研究方法使用热量等值法（thermal contours）来进行识别，并通过模板匹配、分形法（fractal）或小波变换法来进行匹配验证。这种方法只是适合于合作处理控制应用方面，这要求待验证对象主动的配合该系统，例如常见的ATM或者销售终端机（POS）。其精度受影响红外图像轮廓的因素干扰，这包括是否戴眼镜或者是否更换了眼镜，还包括头部相对摄像机的位置、面部或发型的改变等。
Several approaches to feature definition and extraction have been considered. Choice depends on the camera used, the level of subject cooperation, the throughput required, and the price to be paid.
特征定义和提取的一些方法已经得到公认。这些选择取决于所使用摄像机的类型、验证对象的合作程度、系统吞吐量的要求以及所需的花费等。
4.1 Historical Approaches (历史路径)
Metrics Matching. Some current commercial face recognition systems use face metrics obtained from visual images. Those systems admittedly cannot differentiate identical twins, and are inaccurate when the lighting is dim or when it is not uniformly illuminating the face. Even when a face is well lit, differences in angle of view can affect the manual or automatic locating of feature points. Shadows, glint, makeup, and disguise can cause greater errors in locating the feature points and deriving relative distances. Only about a dozen feature points such as inner and outer corners of the eyes, top and bottom points where the ears are connected to the head, centroids of the nostrils, and lowermost tip of the nose can be precisely and repeatedly located if seen in the visual image. Many more feature points can be extracted from IR imagery, by using specific vascular references, such as the branching of particular blood vessels.
度规匹配 当前一些商业性质的人脸识别系统所使用的是可见图像中获得的面部度规(face matrics)来实现的，因此这些系统肯定不能辨别双胞胎，并且当光线较暗或者面部光照不一致时系统辨别也会出错的。既是在光照效果很好的面部图像中，当拍摄角度不同时也会影响特征点的自动或手工定位。阴影、闪光、化妆和伪装等也会对特征点定位和获得相对距离产生更大的误差。在可见光图像的人脸匹配系统中仅有一些特征点例如：眼睛内部和外部的眼角、耳朵连接头部的上下点、鼻孔的矩心、鼻子的最下端等，可以精确的并且可重复的在可见光图像中定位。更多的特征点可以通过具体的血管参照在红外图像中提取出来，例如特殊血管得分支情况等。
Table 1 Range of Face Metrics in Sample （样本的面部度规范围）
Faces scaled to 100 pixels between pupils:
inner eye corners 20 29 内眼角 20 29
outer eye corners 141 178 外眼角 141 178
nose base width 48 62 鼻子基宽 48 62
nosetip to mouth line 28 54 鼻尖至口线 28 54
nose tip to face center 90 118 鼻尖至面孔中心 90 118
face width at nose base 178 215 鼻子基部的面孔宽度
face width at centerline 201 221 中心线处的面孔宽度
face width at mouth line 199 220 口线处的面孔宽度
ear top(s) to face center 160 185 耳朵顶端距离面孔中心
ear bottom(s) to face center 195 230 耳朵底端距离面孔中心
face height above center 122 136 中心线以上的面孔高度
face height below center 143 203 中心线以下的面孔高度
However, there is no large scale test, using either IR or visual imaging, indicating that any person.s set of facial measurements is unique, and often only a few such measurements are available from imagery --- particularly in non-cooperative applications. Face metrics is therefore not considered a robust identification technique.
然而无论是在可见光图像还是红外图像方面都没有大量的测试来证明每个人的一系列面部测量是唯一的，并且通常只有很少的测量数据可以从成像中获得----特别是在非合作系统中。因此这种度规匹配技术不具有健壮性。
Table 1 presents 14 metrics which may be obtained from either IR or visual images. The face center is defined for this purpose as the midpoint between the eyes, located on the line connecting the centers of the pupils. A sample of 100 images in IR and the corresponding 100 visual images were manually analyzed to produce the ranges of measurements shown. All images were scaled to 100 pixels between the pupils. Such metrics can be applied to both visual and IR images as a classification technique to partition the respective databases in order to reduce search time. This can assist rapid comparison of faces-in-the-crowd streaming images against a Watch List of known faces. Partition boundaries need to have sufficient overlap to account for imprecision in feature designation and metric computation
表一列出了可以在可见光图像或者红外图像中获得的12组测量数据。面孔中心在此定义为位于两眼之间的中心，位于连接两眼瞳孔的连线上。通过对由100张红外图像和相对应的100张可见光图像所组成的样本进行手工分析可以获得上述数据。这种测量指标可以用于红外图像或者可见光图像数据库的分级技术，从而减少数据库的查询时间。这样可以加速在流动人群中快速进行人脸查找和匹配。分区界限要有充足的重叠区来解决特征制定和测量计算中不精确的问题.
Template Matching. If thermograms are standardized as to size, and histogram normalized, the areas about the canthi and the inner checks can be compared using template matching with rather good results. On databases of 250 people, a crossover error of 7.5% was obtained by matching three template areas, using older cooled IR cameras having NETD of approximately 0.1o C. At that level of sensitivity, detailed thermal contours are not seen and so shape analysis does not apply. Inexpensive uncooled IR cameras can now provide that level of verification accuracy with a simple processor chip installed in the camera, for a system cost below $5000.
模板匹配: 如果将温谱图根据尺寸大小进行规范化,并将直方图正常化,眼角区域及内部的检验可以使用模板匹配法来获得更好的效果。对于一个250人的数据库，7.5%的交叉误差来源于三个模板区域的匹配，使用较古老的可致冷红外摄像机可以获得大约
面部模板
Elemental Shape Real-Time Matching. With IR cameras having NETD of 0.7o C or less, 100 or more different closed thermal contours are seen in each face. The sets of shapes are unique for each individual, even in the case of identical twins, because they result from the underlying complex network of veins and arteries. Variation in defining the thermal slices from one image to another has the effect of shrinking or enlarging the resulting shapes, while keeping the centroid location and other features of the shapes constant.
基本的形状实时匹配 通过红外摄像机获得
Each nesting of thermal closed contours is called an "elemental shape". Pre-production systems based on elemental shape analysis have been built and have undergone extensive testing with more then 250 persons, and databases of more than 10,000 images.. This has proven the persistence of those features of the facial thermograms which are used for identification and the uniqueness of each person.s thermal image. Facial thermograms for a limited number of subjects have been obtained for periods of 8 to 23 years, demonstrating the required persistence over those durations. Automated IRID using elemental shapes in real time has achieved 96% accuracy for cooperative access control applications.. Studies on pairs of identical twins have demonstrated that they can be separately identified through elemental shape analysis.
每层嵌套的闭合热像图轮廓称之为”基本形状”。基于基本形状分析的预处理系统已经研发出来并且正在对250人的10000张图像的数据库进行广泛的测试。这已经证实了这些面部温谱图中所使用特征的持久性和每个人的面部温谱图的唯一性。少数验证对象的面部问谱图已经获得了8-23年了，表明所要求的持续性超出了该持续时间。使用基本形状的自动红外人脸识别系统可以在写作处理控制应用方面达到96%的准确度。另外研究表明该方法可以将双胞胎辨别出来。
Elemental Shape Post Matching. The totality of shapes in a library of facial thermal images were analyzed. Eigenshape analysis was used to compare characteristics of each shape including: perimeter, area, centroid x and y locations, minimum and maximum chord length through the centroid, standard deviation of that length, minimum and maximum chord length between perimeter points, standard deviation of that length, and area/perimeter. Shapes whose edges are interrupted may need to be ignored when compared against images with different edge effects. As examples: Shapes along the face edge will change when the face is turned. Shapes around the eyes will be affected when glasses are put on.
基本轮廓的后期匹配 在一个人脸热像库中所有图像将被分析。特征形状分析法将用与比较每一个形状的特征，其中包括：边缘、区域、x和y质心的位置、通过质心弦长的最大值和最小值、长度的标准偏差、周边点之间弦长的最大值和最小值、该弦长的标准偏差以及区域、周边等。边缘中断的轮廓当同不同效果的轮廓进行比较时应该忽略掉。例如：当转脸时，脸部边缘的轮廓将会改变。当戴眼镜是眼部周围的轮廓也将会改变。
Thermal Contours of Identical Twins（双胞胎的热像轮廓）
A non-cooperative, non-real-time faces-in-the-crowd version of the IRID system was also built and demonstrated, with more than 100 persons represented in a database of 500 images. Twelve targeted persons were selected, and the database was searched for all appearances by the targeted persons. 98% accuracy was achieved, with no false positives. All the database images were manually selected from IR videotapes obtained at trade shows. Only in-focus, essentially full face images were selected and manually scaled and centered. Processing was not done in real time. The 11-coefficient eigenanalysis required 20 hours on a 486/66 PC.
非协作、非实时在人群中进行的红外人脸识别系统也已经问世并被验证使用，其使用了由100多人的500张图像所组成的数据库。并选出12个目标人物，并从数据库中查询出所有目标人物的图像。可以达到98%的准确率，其中没有出现虚假的判断。所有的数据库图像是在内部预映的红外视频画面中手工选取的。只有清晰的、完整的面部图像被选取出来并且手动改变比例和中心位置。该处理并不是实时进行的。11个特征匹配点的分析过程在486/66 PC上需要20小时。
Figure 3 Variation in Thermal Contours with 10oC Change
改变
Each person.s image was characterized by a set of 11- coefficient polynomials. The difference in eigenspace between any two images was calculated to yield a measurement to which a threshold was applied to make a .match / no match. decision. While those studies proved that IRID can produce high accuracy identification for noncooperative faces-in-the-crowd applications, the calculation techniques are computationally intensive and would need to be hardware-based for real-time use.
每一个人的图像可以由一个含有11个系数的多项式进行表示。任意两个图像在特征空间的区别可以经过计算来开始用于产生一个是否匹配的决策。尽管这些研究表明IRID可以对于非合作、人群辨别应用产生很高的准确率，但所需计算技术是运算密集型的，因此需要基于硬件来完成实时应用。
Figure 4 Sample Corresponding Shapes from Figure 3
（图三中样本所对应的形状）
Wavelet Analysis. Prototype IRID systems for access control were developed by Unisys Corporation under a licensing agreement. The configured systems were designed for unattended cooperative access control. Components included IR camera, Pentium I PC-class computer, face acquisition assembly, keypad, and the enrollment subsystem of a monitor and keyboard. The face acquisition assembly included the mechanisms which move the camera and focus it so that the face is within the field of view. This was the first large scale IRID test to use an uncooled camera --- a BST 320x240 built by Lockheed Martin.
小波分析技术 早期用于存取控制的IRID系统是由Unisys公司在获得许可协议后研发的。其配置系统设计为无人看护的合作性存取控制。其组件包括：红外摄像机、奔腾 I 系列计算机、面部获取集成装置、键区以及有键盘和显示器构成的注册子系统。其中面部获取集成装置包括可以移动摄像机并将脸部包括在视野范围内的机械装置。这是第一次使用非致冷相机（由Lockheed Martin设计的BST 320 x240）进行大规模的IRID测试。
A proprietary wavelet analysis and matching engine from Fast Pattern Recognition was used. Performance exceeded 98.5% accuracy including persons who wore eyeglasses some or all of the time. However, when persons switched glasses, the system error increased appreciably. This error mode needs to be eliminated and the cost of the hardware matching engine needs to be drastically reduced from the $25,000 figure before this approach to IRID can be successfully marketed
在快速模式识别中已经使用了一种专用的小波分析和匹配引擎技术。其性能超过了98.5%的准确性（其中包括有时或者经常佩戴眼镜的人群）。然而，当人们更换眼镜后，系统误差会稍微增加。因此，若想让该技术成功上市，这种错误模式必须应该消除，并且硬件匹配引擎的价格应该从现在的25000美元大大的降价。
4.2 Recent and Current Research IRID systems now under development will utilize lower cost cooled and uncooled infrared cameras, and will perform identification based upon recognition of anatomical structures rather than thermal contours. As the demand for biometric identification systems grows, and as implementations are considered for very large populations, such as national registries and airline passengers, there are related needs for data compaction, rapid comparisons, and adaptable decision levels to meet variable requirements for security and throughput. In response, alternative analysis techniques are under development which take advantage of improved IR camera imagery.
最近和目前的研究 现在正在设计研制红外人脸识别将使用需要更低花费的制冷和非制冷的红外摄像机，并且是基于解剖学结构特征识别而不是热图像轮廓识别。随着生物特征识别系统要求的增高，以及该技术已在越来越多人群的应用，如航空旅客等，这要求数据精简、快速比较以及能够满足可以根据安全需要和吞吐量不同做出适应性决策的水平。因此，基于改进红外摄像技术的红外决策分析技术正在研究之中。
Symmetry Waveforms. Ideally, each human face is bilaterally symmetrical when viewed in either visual or IR imagery. However, no one.s face is truly symmetrical in either spectrum. As camera sensitivity and resolution improve, more subtle asymmetries can be seen. Starting with the hypothesis that the details of each person.s asymmetries are unique to that person, a number of analysis techniques have been developed to exploit those individual variations for identification purposes.
对称波形技术 在理想状况下，无论是可见光图像还是红外图像中每个人脸都是两侧对称的。然而，没有一张人脸在频谱上是真正对称的。随着摄像机灵敏度和分辨率的改进，越来越多的细微的非对称特征显现出来。现在提出这样一个假设：任意一个人的面部不对称特征相对于其他人来说都是唯一的，再该假设下，采用这些个体差异来完成识别目的的许多技术都已经相对成熟。
Six techniques have been investigated. In each case the image is first scaled and histogram normalized. In each case only a vertical swath of the face within the outer corners of the eyes is analyzed, in order to minimize the effects of head rotation. Therefore, these techniques are applicable only to imagery in which both eyes can be seen. Each technique assigns a value to each horizontal line of pixels in the face swath. These values constitute the symmetry waveform.
其中有六项技术已经通过调查研究。在每项技术中都要先将图像规格进行标准化处理。为了尽量减少头部旋转效果，只对包含外眼角的人脸垂直扫描图进行分析。因此该技术只能应用于两眼均可见的图像。每项技术都对人脸条状图中的像素水平线赋一个值。有这些数据来构成对称波形。
The first three techniques analyze the grey scale image and produce a waveform which represents: 1) the total grey scale value to either side of the vertical line bisecting the swath, 2) the x-value at which the total grey values within the swath to either side are equal, 3) the foldover point which provides the smallest cumulative difference. between grey scale values to either side of the point.
前三项技术对灰度级图像进行分析，并产生一个波形图，该波形图用于表明：（1）所有分割面部垂直条出的灰度级数值（2）在总灰度值处的x值（3）折叠点
The fourth technique analyzes thermal contours and locates the balance point between all the contours which cross that horizontal --- with each crossing point given equal weight. The fifth technique is similar but analyzes the vasculature by binarizing but not eroding the image. Larger blood vessels contribute more pixels to the balance calculation.
第四种技术分析热像图轮廓并且在所有穿过水平线的轮廓上设置平衡点---对每一个交叉点设置相同的权值。第五中分析技术很类似但是通过二进制化图像分析了血管分布，而不损坏图像。大量的血管对平衡计算产生了更多的像素。
The sixth technique considers minutiae symmetry, finding the best balance point among minutiae along that horizontal. Due to the relatively small number of minutiae, and the various causes of imprecision in specifying their location, each minutia is represented as a dot which is several pixels in diameter. The size is a function of the expected location error .
第六种技术考虑了细节特征点上的对称，从而在水平线附近的细节特征点上发现最佳的平衡点。由于细节特征点的数量相对较少以及引起指定它们位置不确切的各种原因，每一个细节特征点有直径仅有几个像素的原点来表示，其大小是一个预计定位误差的函数。
Symmetry waveforms can be used for tracking of head rotation, tilt, and tip, and to provide accurate alignment of images. The obtained symmetry waveforms from the latest project are highly encouraging for the subjects considered (23). Each of the six techniques produced unique and repeatable waveforms for the small subject pool used. The results were obtained from an Amber Radiance InSb 256x256 camera with
对称波形技术可以用于追踪头部旋转、倾斜、和末端以及提供准确的图像对齐。在最近项目中得到的对称波形对于研究内容是十分令人鼓舞的。这六项技术中的每一项技术都使用的小型试验群体产生了唯一的、可重复的波形。结果数据来自：Amber Radiance InSb 256x256 camera with
Symmetry Waveform（对称波形）
Face Codes. It may be possible to derive an unique face code repeatedly from each person, eliminating the need for comparing against a database Based upon the degree of variation seen in different persons’ symmetry waveforms, a coding approach was considered which would derive a one-dimensional bar code from a symmetry waveform. Although it seemed straightforward to designate a bar code line for each significant transition in the waveform, it is in fact a sophisticated calculation that is required. Additional improvement is needed in developing related algorithms.
面部编码 或许可以从每个人脸中可重复的提取一组脸部编码，这样就不必要和数据进行比较。根据在不同人脸对称波形图中的变异度（degree of variation），有一种编码方法只值得考虑的，这种方法是从对称波形图中获得一组一维条形码。尽管对波形图中每个明显的过渡设计一个条形码线看起来似乎简单直观，但这是一种必须的先进计算方法。另外还需要改进其他相关的算法。
Figure 6 Bar Code Derived from Symmetry Waveform
（从对称波形图中获得的条形码）
Two-dimensional bar coding of the face has been considered, in which the face is divided into cells and each cell is compared to a library of face segments. The best matching segment is selected and its code used for that cell. Matching is done using Flash Correlation®. While this approach appears promising, further work is needed to establish an optimized library, which should be generated by considering a large database of facial images from the particular camera to be used in a system. A simple binary code can be generated by dividing the standardized face center into cells and indicating those in which minutiae are present.
面部二维条形编码技术也正在受着人们的关注，在该技术中将人脸分割为很多单元，并将这些单元同人脸片段构成的库进行比较匹配。并从中选出最佳的匹配片段对应与该单元。该匹配使用Flash Correlation完成。尽管该方法看起来很有前途，但是由于将来需要建立一个最优库，这将通过在该系统中的使用的专用摄像机提取的面部图像构成的大型数据库检索形成。可以通过在各单元中划分标准化人脸中心并指明其中出现的细节特征点来产生一组简单的二进制编码。
Figure 7 2-D Bar Code from Facial Minutiae
（人脸细节特征点的二维条形码）
More sophisticated encodings also consider the type of minutiae and interconnections. For the databases used to date, which typically included 5000 images of 500 persons, the results from face encoding did not appear to be any more useful that simply down sampling an image that has been deskewed, centered, scaled, and histogram normalized. Resulting thumbnail images that are 32 pixels wide provide accurate positive identification with minimal processing required.
更先进的编码技术还要考虑细节特征点的类型及其相互联系。目前所使用的数据库通常包括500个人的近5000张图像，人脸编码的结果并不是更加有实用而只是简单的向下采样一个已经经过消除倾斜、居中、调整比例及分布图标准化处理的图像。产生的32像素宽的缩略图像提供了精确的经过必要的最小化处理的身份肯定识别。
IR Thumbnail Retains Identifiable Features
（红外缩略图保留了识别特征）
It is expected that an effective thermal face encoding technique will eventually be developed. The coding scheme will take account of head position, and allow for degraded accuracy of identification when only a partial face is seen. Resulting systems will scale up to very large population. Depending on the computational complexity of the processing required, such a capability could allow for rapid throughput with minimal cooperation by the entrant. This technique would also support real time faces-in-the-crowd and digital signature applications.
人们期望的一种有效的红外脸部图像编码技术即将问世。这种编码方案将头部位置考虑在内，并且当只有部分可见人脸时允许降低识别准确度。最终系统会扩展到很大的适用人群。根据处理过程所需的计算复杂性，这种性能可以在参与者的最小合作下实现快速吞吐量。这种方法也将支持实时流动人群人脸识别和数字签名应用。
Thermal vs Visual Minutiae. Another IRID approach focuses on the correlation between IR and visual facial imagery. That capability broadens the security market to include uses where no reference database of thermal images exists. Existing databases of visual images (mug shots, DMV photos, passport photos, company IDs, surveillance camera images) can be automatically matched against images from an IR surveillance camera. The most dramatic application is for automated screening at airports for known terrorists. However, the same system and techniques apply to any surveillance usage; for security at government buildings, courthouses, embassies, foreign corporate offices, and so forth. Although several companies offer facial ID systems which operate with visual imagery alone and claim to provide similar capabilities, they cannot perform accurately when the lighting is dim or when disguises are worn.
热细节特征点及可视细节特征点比较 另外一种红外人脸识别方法致力于研究红外和可见光人脸图像的相互关系。这种性能将安全市场领域扩展到不涉及热像图数据库方面的应用。现有的可见光图像数据库（嫌犯照片、DMV格式照片、护照照片、企业标示、监视器截图等）自动完成和监视器图像的自动比较匹配。最显著的应用是在机场中自动识别已知的恐怖分子。当然，该系统核技术同样可以用于政府机关、法院、大使馆、外交部等部门的监视系统。尽管有些公司中也出现了只使用可见光图像或者声称拥有类似功能人脸身份验证系统，但是当光线较弱或者化妆后其精确度将大大下降。
The IR/visual correlation capability extends the markets for IRID beyond security into communication process control applications. In particular, use of a dualband IR/Visual camera enables significant bandwidth compression of "talking head" video. A system user or communications service provider can achieve high compression ratios by transmitting only a baseline visual facial image, followed by movement vectors for those thermal minutiae which are involved in changes of expression or speech. The animated face is then reconstructed at the receiving end. Applications for this technology include: videotelephone, video email, videoconferencing, news broadcasts, and electronic commerce. Together, these applications represent a sizable and growing percentage of the overall transmission of internet and telephone messages. (基于红外/可见光相关联该的IRID技术在通信处理控制领域的应用)
这种红外/可见光相关联的性能将IRID市场从安全领域扩展到通信处理控制领域。尤其是双频IR\Visual摄像机的使用明显实现了对“Talking/Head“视频带宽的压缩。系统使用者或者通讯服务提供者通过只传输可见光人脸图像可以获得很高的压缩率，另外这些热像图的细节特征的运动矢量也会随着表情或者讲话的改变而改变，生动的面部就会在接收端重新绘制出来。
IR/Visual correlation also extends the markets for IRID to health care. There is a growing market for home diagnostic devices, and infrared cameras linked to diagnostic software are destined to become as common as thermometers in the home medicine cabinet. We anticipate that routine infrared monitoring of elderly patients on heart and arthritis medications, and postsurgical care of wounds or infections will be among the first applications for home-based diagnostic on-line systems. Household accidents, including falls and burns and insect bites, will also be imaged with a home infrared camera which transmits live video to an HMO（卫生维护组织） diagnostic center for realtime advice and preadmission approval to emergency centers. Future diagnostic uses for infrared and for hyperspectral imaging will be tremendously varied; in homes, nursing homes, clinics, hospitals, ambulances, police cruisers, gymnasiums, and basically anywhere there is currently a first aid kit. Those applications, however, are outside the scope of this paper, which deals only with the identification aspects of infrared imaging. （基于红外/可见光相关联该的IRID技术卫生保健领域的应用）
Identification is, however, an important aspect of medical care. It includes issues of patient privacy（保护患者隐私）and who has access to patient data. It includes assuring that the proper patient receives the proper treatment. It includes procedures for computerized patient recordkeeping（将患者记录存入电脑）. It also includes the issue of insurance and medicare fraud（防止骗保）. Those issues can all be addressed by several different security approaches, including other biometrics. In every case the basics are the same: each person is positively identified prior to any service being performed, or any access granted. There is another medical identification requirement, however, for which IRID is uniquely applicable. （验证技术在医疗处理方面的重要性）
The same cardiovascular network(心血管网络), including blood vessels and other elements of the vascular and lymph systems（淋巴系统）, which permeates the face area is similarly distributed throughout the body. This provides the capability for identifying any body part just as it provides for identifying the face. More importantly, it provides a network of approximately 2400 minutiae（大约2400个细节特征点） whose locations and interconnections can repeatedly be determined passively and from a remote imager（远程成像仪）. These 2400 points serve as anchor points（锚点） to facilitate morphing an image to a standard image, or to another image of the same type from the same patient. Since infrared cameras are passive, and are becoming inexpensive compared to other medical imaging devices such as a sonogram（声波图）or X-ray, an IR camera can be used in conjunction with other imaging systems at minimal additional cost（成本低）. This provides the capability to annotate the IR minutiae onto other types of medical images; thereby facilitating sensor fusion among a variety of diagnostic medical system.
An even simpler application is the automated comparison of same-modality images taken at different times（不同时间拍摄的相同模式图像的自动匹配比较）. The analysis system would determine whether a particular medical condition（个别医疗状况） was improving or static or regressing. The thermal minutiae（热详细节特征点） would be used for automatic alignment（自动校准） of the two images, which could have been taken using different imaging systems, which produce different resolution（分辨率） images, from different distances and angles, and with different areas of view. With the collection of a large volume of annotated imagery and associated patient histories, we predict that automated libraries will be generated. A given patient’s current medical images could then be automatically compared not only with his own earlier images, but also to standardized images representing composites formed from other patients’ data in the library. That would enable automatic determination(自动测定) for example that not only is a surgical patient improving since his last image was taken; but that with a confidence level of 95% his progress is most similar to a patient group which had a 90% survival rate ten years after the same surgery. （另一个应用领域：医疗图像的匹配è诊断）
Perhaps one of the good outcomes of managed health care will be the opportunity to better manage health care information（更好的管理卫生保健信息）. Use of the Standardized Infrared Minuties Coordinate System (SIMCOS) technique is expected to facilitate improvements in medical informatics, by providing a structure for development of standardized image libraries and automated comparison techniques. （应用的优点）
4.3 Anticipated Performance of Future IRID Systems(未来IRID系统的预期性能)
The cameras used during the twenty year development of IRID included a variety of detector materials and cooling techniques, array sizes, and optics with different resulting resolution and sensitivity. However, there has been a steady decline in prices and improvement in performance. （过去的20年里各种仪器设施降价、性能提升）The decline was noticeably accelerated by DARPA’s funding during the
Dual-mode "look at / look for" IRID systems (双模式的“查看/寻找“IRID系统)are envisioned which will first identify an authorized user（特许用户） and then provide him with access and IR viewing capability. Such systems would be deployed on weapons（武器装备）, in vehicles, and associated with remote surveillance operations. They provide dual use of a single IR imager, thereby offering reduced overall weight and cost. （未来的双模式的“查看/寻找“IRID系统其特点及应用）
5. Areas of Application for IRID （IRID的应用领域）
Research to date has investigated the various IR feature extraction and analysis techniques for human identification in five application areas. （现有的红外特征提取和分析技术的五个应用领域）。
5.1 Access Control. （进出口管制）
IRID systems for access control aim at insuring the identity and authorization of persons (获得认证和授权)seeking entry to physical locations or in transactions for which only an initial identification(初始鉴定) is required. The IRID system can be stand-alone or be integrated into other systems. （使用目的）
Prototype IRID systems for access control were developed by Unisys Corporation under a licensing agreement. The configured systems were designed for unattended cooperative access control（原始的IRID系统的特点：无人看护、配合操作）. Components included IR camera, Pentium I PC-class computer, face acquisition assembly, keypad, and the enrollment subsystem of a monitor and keyboard. The face acquisition assembly includes the mechanisms which move the camera and focus it so that the subject.s face is within the field of view. The primary cost component of the system is the infrared camera（主要组件是红外摄像机）. The camera in the current system was designed and built by Lockheed Martin for Unisys. It uses an uncooled BST array and produces images of 320 x 240 pixels in the 8-12 micron band. The stand-alone access control system version of the IRID system was designed to sell for $25,000, and to offer a false positive rate （伪正率）of less than 1% and false negative rate（伪误率） of less than 2% for single access attempts. Improved systems now under development will incorporate various hardware, software, and procedural modifications. which are expected to improve system accuracy while reducing the price. (IRID系统的组成机器各组件的性能和发展前景)
5.2. Computer Security.(计算机安全)
IRID offers exceptional features for real-time computer security.(实时计算机安全) IRID continually re-verifies the identity of the system user, and confirms his authorization level prior to each action he performs. The continuity verification（优点：实时持续验证） uses the patented FlashCorrelation® technique which permits normal head and body movements within the field of view of the camera. However, the system automatically blanks the screen, locks the keyboard and peripheral devices if the user moves out of the field of view of the camera, or does not look at the monitor or interact with the system for a programmable period of time. （特点）
The IRID computer security system provides positive continuous biometric identification of persons logging-on and interacting with protected networks. The infrared vascular pattern provides a digital signature which can be used to tag files created or accessed. Once logged-on, the system does not interfere with the user other than to control activities based upon prior authorizations. Onlookers can be detected and their presence used to blank the screen and otherwise temporarily disable the system.（特点：计算机操作必须连续性而不允许中断，用户必须在摄像机视角范围之内） Movement of the user out of the field of view of the IR camera also blanks the screen and temporarily disables the keyboard and peripherals. Return to within the field of view automatically re-enables the system, although prolonged absence will cause automatic log-off.
5.3. Smart Surveillance.（智能监控）
IR imaging has significant advantages for "smart" surveillance, which has been coined to refer to automatically screening persons seen in surveillance operations against a watch list of persons who are to receive special attention(自动监视是否在监视屏幕中出现了在监视名单中的人). The watch list will generally include persons who are to be denied access to a facility, or whose presence is to be reported to authorities. However, a watch list could also contain persons who are to be given preferential treatment or access, such as undercover police.（监视名单包括的人物可以给予不同的待遇）
The IRID smart surveillance systems will extend the usage of IR surveillance to automated identification of persons whose IR facial image has not been previously collected, but whose visible facial image is available. （只有可见光图像而无红外图像时也可以辨别）These systems will utilize coincident features which appear in both imaging modes plus anatomical features seen in the IR image. IRID Smart Surveillance Systems will provide the ability to automatically surveil an area in the dark（在黑暗中）, find each face （发现人脸）within the scanned area, and match it against a Watch List of visible images.
Recurring offenders and locations with recurring criminal incidents are prime candidates for installation of Smart Surveillance systems. Parole or probation of repeat offenders（假释或缓刑的惯犯） can include prohibitions against visiting hot spot areas. Smart Surveillance technology provides the means to verify compliance in a cost-effect way. Counterterrorism（报复性恐怖组织） will be aided by the ability to automatically check each person entering a monitored location such as an airport, customs area, or courthouse against a Watch List of known or suspected terrorists.(监视对象类型及场所)à已知的监视对象
Crime prevention and crime solving (犯罪预防和案件侦破)can also be aided by the automated comparison of IR facial images collected at different locations or at different times even when the subjects are not included in any current Watch List. The fact that an unknown person is visiting certain sites or a particular sequence of sites, or that he appears on-site immediately following an incident, or that he is seen to meet with persons on a Watch List may provide important linkage data in predicting criminal activity or correctly assigning responsibility after it occurs. （对未知的监视对象的处理）
5.4. Talking Head Video Compression.(Talk-Head 视频压缩)
Change in facial expression and the action of speech(面部表情和讲话方式的改变) cause movements in affected areas of the face, particularly the lips but also the eye, chin, forehead, and cheek areas. Encoding of facial expressions and facial movements during speech is currently being studied for bandwidth reduction in the transmission of "talking head" video for applications such as videophone, videoconferencing, video email, synthetic speech, and face animation. (图像压缩的应用领域)The intent is to transmit a baseline image followed by encoded changes to that image, with reconstitution of the animated face at the receiving end. This process offers significant bandwidth reduction, （缺点）but may produce imagery in which the talking face seems stiff and unnatural or does not appear to be synchronized with the audio, giving the unacceptable look of a dubbed foreign film.（指传输编码后的人脸基线编码，然后再接收端还原人脸，主要是可以减少带宽）
All previous studies involved modeling the facial movements（人脸运动建模） based upon the relocation of certain observable points of the face（人脸参照点）, such as the corners of the mouth. The various models differ in the extent to which they consider the underlying facial muscles and nerves. （人脸参照点较少，分布特点）There are few observable reference points on a generalized face, especially under uncontrolled lighting conditions. In particular, there are no observable reference points in the cheek areas, and none in the forehead area except possibly skin creases. When the talking head is that of a dark skinned person, the reconstructed image may show further degradation of subtle facial changes. （研究内容，影响参照点的因素）
Use of a dualband IR/Visual camera at the transmission end offers the potential for marked improvements. Infrared minutiae are more numerous than visible markers and are present throughout the face, including areas of the cheeks and forehead and chin where no visible minutiae may be present. Therefore, modeling of the movements of infrared minutiae can provide finer detailed replication of expressions and speech than modeling based upon visual references.( 双频红外/视觉相机的使用，红外细节特征点多于可见光图像上的细节特征点)
By processing sequences of images taken from known expressions and/or known speech elements, a sequence of movements of infrared minutiae can be extracted which correspond to that expression or speech element for that person or for persons in general. （红外特征点的提取） Subsequently, when the same sequence of movements of infrared minutiae is seen, it can be inferred that the person is displaying the same expression or speech element as during the initial sequence. This enables the automated determination of expression or speech, （表情及话语识别）allowing for compression of transmitted video in conjunction with audio. （意义）The combination may offer additional composite compression and improved synchronization. The same basic technique can also be used to create a dictionary of facial expressions and speech elements for use in animation of a synthetic face.
5.5. Health Care. （卫生保健）
Increasing use of medical imaging techniques （医学图像技术）is producing an enormous amount of medical history data （病史数据）for each patient. Medical diagnosis and treatment increasingly involves imaging from various sensor modalities, such as: x-ray, CAT scan, MRI, sonograms, and mammograms. As a person grows older, moves, changes doctors, experiences various medical events, changes employer and insurance company and perhaps his or her name, the task of maintaining a complete and accurate medical history becomes increasingly complex. The volume of information per patient, coupled with the desire to reduce medical costs, has led to a goal of computerized patient records. The growing use of HMOs and telemedicine requires methods by which physicians can rapidly review and assess computerized medical records containing image data and written comments generated by various medical personnel on patients they may have never seen.
A primary requirement in establishing computerized records is for a method of automated patient identification.（自动的患者识别方法） It is essential that the correct diagnosis and treatment are performed on the correct patient. This includes emergency patients who are unconscious, mentally incompetent, intoxicated, or otherwise unable to identify themselves.（适用对象） All computerized medical data must be correctly annotated with the patient.s ID and filed in the correct patient record. Automated identification is particularly important in large databases where they may be multiple persons with the same name, and in telemedicine applications where the patient, his medical records, the primary and consulting physicians may all be in different locations. Aside from accidental errors associated with human and system mistakes, there are intentional errors created for reasons of insurance fraud, in which imagery or other data submitted with claims is not from the person represented, or in which multiple claims are filed under multiple identities for the same person. The need exists to improve the accuracy of associating medical imagery and other data with the identity of the person from whom it was derived.（在卫生保健领域使用IRID的必要性）
A primary requirement for medical data exploitation is for automated comparison(自动比对) of images taken from the same individual, but at different times, under different conditions, and perhaps using different sensors. This type of sensor fusion can utilize a standard set of reference points （标准特征点）on the human body which will accommodate changes in age, weight and health status. Since the reference points（参照点） may need to be frequently determined, their collection must pose no risk to the individual. A technique is needed which annotates standard reference and registration points on any and all medical imaging modalities, and which can be used to standardize medical library files in order to compare different imagery from the same person or from different persons.
Work is proceeding on a Standardized Infrared Minutiae Correlation System (SIMCOS) for annotation of standardized thermal minutiae onto medical images. The minutiae serve to identify the patient, align different images, and unlock access into the patient.s computerized files. （主要应用）The SIMCOS approach is analogous to the use of fingerprints for automated identification of persons; （缺点）however, it utilizes all or any part of the body.s surface, and requires no contact or cooperation with the patient. （适用对象）Therefore, it is particularly suited for dealing with noncooperative patients such as the aged, infants, unconscious, mentally handicapped, or drug and alcohol impaired. Minutiae patterns can serve as a security access code for opening the specific patient.s electronic files
Vasculature seen in IR Imagery（红外图像中的脉管系统）
Current commercial IR cameras do not have the resolution to provide direct viewing of vasculature over the entire face in a single frame. (当前商业红外摄像机不能提供脸部直观的脉管系统)However, as Figure 9 indicates , partial face views are highly detailed and can be mosaiced to cover the entire face or other body area. A microscope lens allows for viewing the specific influence of an underlying blood vessel on the skin and hair above. （未来红外摄像机的发展对卫生保健领域的应用）Future IR camera developments for medical applications may renew interest in linear array imagers. It is expected that hyperspectral linear arrays（高广谱线性阵列） will be developed for whole-body scanning. The same minutiae extraction and SIMCOS standardization technique will be applicable to the resulting imagery.
Standardizing database images facilitates comparison of imagery during growth from childhood to adulthood, compilation of medical libraries of images from large numbers of people, and automated comparisons of current imagery against the vast libraries for diagnostic purposes. Standardization to a common set of thermal minutiae would also provide a common reference for comparing images obtained from different sensors which produce different resolution images. There are many potential applications for medical image fusion, and for automated positioning of patients relative to diagnostic, surgical, and treatment equipment. One example follows.（标准化数据库图象的优点及其应用）
Landmarks on nuclear medicine studies.（核医学研究的里程碑）
Certain studies, such as for myocardial perfusion（心肌灌注）, require two or more studies be performed on a given patient at two different times. Precise repositioning is difficult due to the many relevant factors which are not known in advance, such as orientation of the heart within the body and size of the various portions of the heart.
The requirement for repositioning may therefore lead to successive manual repositionings of the camera and patient in order to repeat as closely as possible the earlier study. Using SIMCOS, standardized thermal minutiae can be used for universal positioning of all patients, or for specific repetition of a given patient.s positioning. The thermal minutiae can be annotated on the resulting imagery. The SIMCOS technique is expected to be more precise and repeatable than current techniques for identifying the long axis of the left ventricle from nuclear medicine SPECT data.(使用SIMCOS的优点)
6. Other Future Applications （其他的未来应用）
The systems described to this point have all been installed on fixed or mobile platforms from which they identified whatever person was within their field of view. There are also significant applications for miniature IRID systems which would be integrated into worn computer systems and dedicated to watching the single specific wearer. Design of the bodyworn IRID system (随身IRID系统)would likely involve a Bluetooth-compliant communications capability (蓝牙通信功能)to securely broadcast in a limited range the identity and status of the wearer.
Transmitted information could include the wearer’s level of attention, point of attention, level of fatigue, indications of stress, facial expression, and whether or not he is talking. This information is derived from a combination of thermal minutiae micromovement analysis and monitoring of such classic observables as blink rate and eye fixation patterns, sweat, and head nod.
System design involves using a bodyworn computer, with the remote sensor head of an IR camera mounted along with a microphone on a headset. The rest of the IR camera would be worn along with a computer on a belt or backpack. Initial systems would continually, or on demand, broadcast the identity of the person. They would also monitor and transmit his attention, stress, and fatigue levels. Those are all capabilities which are within the capability of current IR imagers and algorithms but have not yet been prototyped.
Future systems would use IR facial monitoring to enhance speech recognition(增强语音识别), particularly for use in noisy environments. They would also incorporate features to perk-up the wearer’s attention, and require additional confirmation of activities initiated when the wearer’s stress or fatigue levels were above threshold. Incorporating these additional features will require extensive further research.
