Font Size: a A A

Research On Pedestrian Detection And Tracking Technology Based On Video Sequence

Posted on:2022-09-06Degree:MasterType:Thesis
Country:ChinaCandidate:K WangFull Text:PDF
GTID:2518306335486024Subject:Computer technology
Abstract/Summary:PDF Full Text Request
In recent years,with the rapid development of information technology,pedestrian tracking has received more and more attention from experts at home and abroad.Pedestrian tracking is mainly divided into two parts: pedestrian detection and pedestrian tracking.Pedestrian tracking uses technologies such as pedestrian detection and recognition to determine the trajectory of pedestrians in the video.Pedestrian tracking has broad market prospects and has extremely important applications in autonomous driving,intelligent transportation,and military drones.The rise of deep learning,especially the large-scale application of convolutional neural networks in the field of computer vision,has further improved the performance of pedestrian tracking algorithms combined with deep learning,thereby promoting the research of pedestrian tracking.Aiming at the problem that pedestrian detection algorithms based on convolutional neural networks cannot coordinate detection accuracy and detection speed,this paper proposes YOLO v3 based on convolutional neural networks as a pedestrian detection model.The detection effect of YOLO v3 is in the forefront of current target detection algorithms.The YOLO v3 algorithm treats the object detection problem as a regression problem,and uses the convolutional neural network structure to predict the target position and category probability of the input image,and at the same time further improves the speed of pedestrian detection.In order to enable YOLO v3 to further adapt to the realtime requirements of pedestrian tracking,this paper also pruned the model of YOLO v3 to improve the speed of pedestrian detection while maintaining the original accuracy rate.This paper proposes a pedestrian tracking algorithm combined with a pedestrian reidentification algorithm.The research idea of pedestrian tracking is to combine deep features with Kalman filtering,Hungarian algorithm,etc.Although the algorithm has high efficiency,it will produce frequent ID switching,especially in video pedestrian tracking.Scenes such as occlusion can easily cause the pedestrian tracking algorithm to track the wrong object.Therefore,it is necessary to combine pedestrian re-recognition to correct the possible errors in pedestrian tracking.By extracting features of pedestrian targets tracked in the front and rear frames of the video,and then performing cosine calculation on the extracted features through pedestrian re-recognition,the matching similarity is obtained.Correct the wrong trajectory of pedestrian tracking.The experimental results show that the pedestrian tracking algorithm combined with pedestrian re-identification has a significant improvement in tracking effect compared to the previous algorithm.
Keywords/Search Tags:Pedestrian Detection, Pedestrian Tracking, Convolutional Neural Network, Pedestrian re-identification
PDF Full Text Request
Related items