COMP9517 Week7 Tracking

https://echo360.org.au/lesson/9750cbfe-6500-42f7-b024-45c5cf2a69b0/classroom#sortDirection=desc

https://webcms3.cse.unsw.edu.au/static/uploads/course/COMP9517/20T2/bca52a9037053fc7f77869b5c36edef5830da7920d455dce479c282c9ba12431/COMP9517_20T2W7_Part_1-1_Motion.pdf

 

Tracking :

 

1. Bayesian inference Using probabilistic models to perform tracking

　　1). • A moving object has a state which evolves over time; Random variable Xi can contain any quantities of interest (position, velocity, acceleration, shape, intensity, colour, …)

　　2)   The state is measured at each time point ; Random variable Yi  measurements are typical features computed from the images

　　3)  贝叶斯预测三步

　　　　

　　　　（1）Prediction 基于前i-1步的measurements预测第i步的state 

　　　　　　

 

 

　　　　（2）Assiciation：如果有不同的objects 则需要选择不同的 object-related measurements (这里预测单个object 没有这一步)

　　　　  (3)   Correction： 用正确的Yi更新之前的预测

　　　　　　　　

 

 

 

 

 

 　　4） Current state only depends on the immediate past; Measurements depend only on the current state

　　　　隐马尔可夫模型

　　　　

 

 

 

 

 

2.• Kalman filtering Using linear model assumptions for tracking

　　　　假设object线性运用,noise成高斯分布

　　　

 

 　　

 

 

　　　　

 

 

 

 

3.• Particle filtering Using nonlinear models for tracking