Research On Pedestrian Detection And Tracking In Videos

Pedestrian detection and tracking is to detect and identify pedestrian from video sequences,and track the target of interest in order to get its motion and state information.It is an important foundation for subsequent research such as behavior understanding,event detection.As an essential research in the field of computer vision,it has been widely used in intelligent visual surveillance,intelligent vehicle,group analysis and management,and public space design,etc.Although the research on pedestrian detection and tracking has made a lot of progress,due to the difference among the pedestrian targets in the static and dynamic appearance,and the variety of different postures,perspectives and distances from the camera,etc.,this research work is still a challenging problem.This paper focuses on the research of pedestrian target detection and tracking in video sequences.The main achievements include:?.An improved pedestrian detection algorithm based on the non-maximum suppression is presented.Based on the pedestrian detection algorithm of HOG feature and SVM classifier,a non-maximum suppression method is introduced to correct the detection results for the problem of candidate box overlapped redundancy,which achieves the accurate detection of pedestrian targets under various scenarios.First,a sliding window is applied to obtain the classification window from the image to be detected,and the HOG features of each window are calculated.Then,the HOG features of each classification window are input into the trained SVM classifier to get the classification results.Finally,the non-maximum suppression algorithm is used to preserve the candidate boxes with high confidence in the detection result while remove the redundant ones,thus the accurate detection result can be obtained.The experiment shows that the improved algorithm can achieve better detection results in multiple scenes,which is more accurate than the traditional detection algorithm.?.An improved particle filtering tracking algorithm based on sparse representation is proposed.The algorithm is robust for tracking pedestrian with abrupt motion,and solves the particle diversity problem of the particle filtering tracking algorithm.First,the sparse representation is used to compute particle weights by considering the fact that the weights are sparse when the object moves abruptly,so the effective reconstruction of interested objects can be gained and the potential object region can be predicted more precisely.Then,a resampling algorithm based on nonlinear sorting strategy is introduced.In the resampling process,more effective particles are retained,which solves the problem of particle diversity impoverishment caused by traditional resampling methods,and has a more robust tracking performance for abrupt motion targets.Experimental results verify the effectiveness of the improved algorithm for tracking pedestrians and other moving targets.The algorithm achieves accurate tracking results in video sequences of multiple scenes.