Research On The Algorithm For Human Detection And Tracking On Camera

With the rapid development of computer technology, target detection and object tracking of intelligent video surveillance has become a hot topic in the field. Object tracking is an important research topic in pattern recognition, computer vision, image processing and other areas. It has a wide range of applications in video surveillance, military, intelligent traffic. This paper focused on some key technologies about human detection and tracking based on the effective extraction of human movement in a single camera.This paper proposed two different tracking methods based on Bayesian tracking and based on probability density estimation. The main work of the dissertation included:1,The dissertation aimed at the problems of the interested human region extraction. Firstly, it detected the human body region based on adaptive Gaussian mixture background modeling method. Then it removed the shadow produced by the light conditions. At last, appropriate morphological operator was used.2,The paper proposed an enhanced particle filter tracking algorithm based on Bayesian estimation to achieve a single camera tracking. The traditional particle filter method was improved. The Kalman filter was added to the traditional particle filter. Firstly, it obtained a certain number of particles from the prior probability of particle filters in the dissertation. Then it updated the state mean and covariance based on Kalman filtering. And then it got Gaussian distribution using the updated mean and covariance. New particle set from the distribution was sampled and particle filtering was performed. The dissertation has achieved the human tracking through the iteration of Kalman filter and particle filter.3,The dissertation proposed a method to track multiple human bodies in a single camera based on probability density estimation. Firstly, we cluster the color information on the surface of the body using the mean shift algorithm and segment object into blocks according to the clustering results. Next, we model the color information based on kernel density estimation in each color block. At the same time, we establish the motion model based on the location information of the human body in the continuous two frames. The algorithm tracks multiple human bodies in a single camera by calaulating the maximum joint probability of the colors and the positions between the storage model and the foreground obtained by subtracting the background.