Research On Video-based Object Tracking And Crowd Density Estimation

With the rapid population growth and economic growth in recent decades, while urbanization bring convenience to people’s life but also bring security problems. While social security incidents occur more and more frequently, the demand for security monitoring is growing. Benefit to the rapid development of image processing technology, intelligent video surveillance technology, as an effective approach, has attracted more and more attention. Video-based object tracking and crowd density estimation are the fundamental technology of behavior analysis, recognition and intelligent video surveillance system. This paper systematically summarized on previous related works, and made a research on video-based object tracking and crowd density estimation technology. The details are as follows:1). This paper first points out two disadvantages of particle filtering algorithm: first, the traditional importance resampling can’t handle the situation when the object posterior probability density is multi-model; second, most algorithm assume that the object motion is smooth, however, the assumption is unrealistic. This paper proposed adaptive resampling scheme and introduced variable structure multiple model estimation method to solve these problems.2). For multi-object tracking application, we point out some disadvantages of applying JPDA algorithm into multi-object tracking:the observation just contains position information and it’s not suitable without applying image information in video-based tracking. To solve this problem, this paper extended the observation information by introducing image features, which improved the accuracy of the association probability matrix, and also the tracking results.3). For crowd density estimation problem, this paper first analysis advantages and disadvantages of pixel-based method and texture analysis method. After elaborated their complementary in various density level situation, this paper proposed feature fusion method, which combined pixelbased and texture feature and designed corresponding fusion strategy, to estimate crowd density more robustly.4). For every every topic mentioned above, various authoritative video library in related fields were collected. Plentiful simulations were made and follows detailed analysis and comparison.