Technology Of Target Detection And Multiple Object Tracking In Intelligent Video Surveillance

Researching on intelligent video surveillance has attracted lots of attentions.The algorithm of foreground extraction, the algorithm of intelligence target recognition and the algorithm of objects tracking algorithm are hot topics. This paper improves the previous work,and tries to overcome the weakness of related algorithms. The main contribution of this paper is summarized as follows:1) This paper improves the contents of Vibe,and makes contributes to this foreground extraction algorithm.We compared the improved Vibe algorithm with the original Vibe algorithm and Mixed-Gaussian algorithm in Change Detection datasets, the result proves the improved Vibe algorithm performs better than the others.Firstly,compared with the original Vibe algorithm, the recall rate gets6%to11%percentage higher than before,with almost no loss on the accuracy rate and computation speeds.Secondly, compared with Mixed-Gaussian algorithm,the improved Vibe algorithm performs much better in accuracy rate, recall rate and computation speeds.2) This paper presents a Mixed-Channel Integral Histogram method,which could compute any histograms of the images in only O(1) time complexity.This method avoids a lot of repetitive computations, especially in the superimposed areas.This method computes HSV color integral histogram in both original image and foreground image.3) This paper presents a Combined Decision Forecasting And Tracking algorithm which is based on the theory of Compressed Tracking algorithm and Mixed-Channel Integral Histogram theory. During the sampling period, this method computes the histogram of each sample, and finds the optimal positive sample and optimal negative sample at the same time. The final decision is made by combining the Bayes classifier of Compressed Tracking algorithm and the scores of histogram similarity in both of the optimal positive sample and the optimal negative sample. The final result performs better than before. This paper presents a multiple objects tracking framework using the target detection algorithm and Combined Decision Forecasting And Tracking algorithm mentioned before. An optimal queue is set up for the best positive sample features.Then the disappearance objects can be recognized thought the KNN classifier algorithm.Experiment results prove the effectiveness of this method.As a result, this paper offers a better foreground extraction algorithm and a better tracking method, which make contribute to our video surveillance system.