Research On The Technology Of Moving Target Tracking And Recognition In Video Sequence

Moving target tracking and recognition is a key technology in the field of computer vision, which combines image recognition,artificial intelligence, image processing and other subjects and used widely in the modern military field, intelligent traffic management,safety system and intelligent simulation system.On the basis of the tracking and recognition algorithms, the paper proposes some improved methods, and prove the accuracy and effectiveness of the methods by experiments.The main work of this paper is as follows:First of all, in terms of moving target tracking, the paper mainly studied corrected background weighted histogram based on mean shift(CBWH), when the speed of the target changes, the algorithm maybe fail, aiming at the question, this paper proposes an improved CBWH algorithm, namely calculating two iterative during tracking, first time is using mean shift and calculating the similarity, second time using12--iicentercenter as the center to calculate the similarity. Compared two similarity coefficients,choosing the larger one as the new center. Based on the different scenarios of the experimental results show that: the improved algorithm has better accuracy than original algorithm.Secondly, in terms of target recognition, the paper studied bag-of-word model recognition algorithm, through a lot of experiments and found the target recognition accuracy has a great relationship with the neighborhood sub-block width in the SIFT algorithm and the number of vocabulary. Using the 5 cross validation on the two key coefficients, and obtains the optimal combination of the two key coefficients on five-fold cross validation. The experiment results show that: when the two key coefficients achieve the optimal combination,the algorithm proposed in the paper has better accuracy than traditional bag-of-word model algorithm.