The Research Of Single/multiple Target Tracking Algorithm Based On Video Sequence

Video target tracking is one of the hot and focal issues in the field of computer vision, which combines many advanced technologies such as image processing, artificial intelligence, automatic control and pattern recognition. Video target tracking has been widely used in human-computer interaction, intelligent video surveillance, robot navigation, and medical diagnostic fields. So far many target tracking methods have been proposed, but designing a robust, stable target tracking method is still a challenging task because of diversity of target’s feature, change of target’s appearance and background, inter-object occlusions.Video target tracking can be divided into single-target and multi-target tracking according to the number of targets. According to whether the target is automatically initialized it can be divided into autonomic tracking and semi-autonomic tracking. In this paper, we study on the semi-automatic single target tracking with fixed background and the automatic multi-person tracking problem. The major work and research are like the following:1. The on-line boosting algorithm regard tracking problem as a binary classification problem. With its on-line learning ability, it is adaptive to objects of different appearance. However, this method can lead to tracking failure due to accumulation of error features when the object is seriously or completely obscured. A new algorithm combining particle filter and on-line boosting is proposed to overcome the shortcomings of traditional on-line boosting algorithm. The confidence of the particles’ region is set to be the weight of the particles and the object tracking situations under serious occlusions was solved.2. This paper presents a moving pedestrian detection method based on Vibe-Hog, which combines the target detection algorithm based on moving and statistical learning. Firstly, morphological filtering and neighboring region merging strategy is used to remove the effect of noise in moving region which is extracted by vibe method. Secondly, the pedestrian detection classifier(HOG+SVM) is used on the extracted moving region to extract the pedestrian. This algorithm reduces the redundancy detection area, improve the pedestrian detection speed.3. Prevent a multi-target tracking algorithm based on detection and Kalman-Meanshift tracking. In Kalman-Meanshift tracking framework, select two features to describe the target through online feature selection. To improve the target tracking ability under complex environment, we association of the detection results and tracking results, and fusion the matching detection values and tracking values as the tracking results...