Opencv Q&A_6

2022/03/12-2022/03/13

基于YOLO3的物体识别

代码

import cv2 as cv
import numpy as np

whT = 320
confidence_thres = 0.5
NMS_thres = 0.2

## 模型配置文件和权重文件路径
modelConfig = 'D:\work\\automation\Skill\Python\pythonWORK\cv\yolo3\yolov3.cfg'
modelWeight = 'D:\work\\automation\Skill\Python\pythonWORK\cv\yolo3\yolov3.weights'

with open('D:\work\\automation\Skill\Python\pythonWORK\cv\yolo3\coco.names') as f:
    className = f.read()
    className = list(className.split('\n'))

cap = cv.VideoCapture(0)

net = cv.dnn.readNetFromDarknet(modelConfig,modelWeight)    #使用darknet框架读取模型
net.setPreferableBackend(cv.dnn.DNN_BACKEND_OPENCV)     #设置计算后台
net.setPreferableTarget(cv.dnn.DNN_TARGET_CPU)      #设置目标设备(用什么计算)

def findObject(outputs_,img):
    hT,wT,CT = img.shape    #图片格式(h,w,3)
    bbox = []
    classIds = []
    confidence = []

    for output in outputs_:     #遍历每个预测
        for line in output:     #遍历每一层
            classId = np.argmax(line[5:])       #获得最大confidenct的序号
            if line[5:][classId] > confidence_thres:    #前五位数值不参与判断
                w, h = int(line[2] * wT), int(line[3] * hT)  # 第三四位数值为宽高值(百分比)
                x, y = int(line[0] * wT - w / 2), int(line[1] * hT - h / 2)  # 第一二位数值为中心点的坐标(百分比)
                bbox.append([x,y,w,h])
                classIds.append(classId)
                confidence.append(float(line[5:][classId]))
    indice = cv.dnn.NMSBoxes(bbox,confidence,confidence_thres,NMS_thres)

    for i in indice:
        ## NMS找出的识别索引
        i = i[0]
        box = bbox[i]
        classId = classIds[i]
        confid = confidence[i]

        cv.rectangle(img,(box[0],box[1]),(box[0]+box[2],box[1]+box[3]),(0,255,0),2)
        cv.putText(img,className[classId],(box[0],box[1]-10),cv.FONT_HERSHEY_SIMPLEX,1,(0,255,0),2)
        cv.putText(img, str(round(confid*100,2)) +'%', (box[0] + 150, box[1] - 10), cv.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)

while True:
    flag,frame = cap.read()
    blob = cv.dnn.blobFromImage(frame,1/255,(whT,whT),[0,0,0],1,crop=None)
    net.setInput(blob)
    layerNames = net.getLayerNames()    #获得每个层的名称
    outputNames = [layerNames[i[0]-1] for i in net.getUnconnectedOutLayers()]   #getUnconnectedOutLayers()：获得非连接层(即预测输出层)
    outputs = net.forward(outputNames)  #前向传播三个预测值输出
    findObject(outputs,frame)


    cv.imshow('img',frame)
    cv.waitKey(1)