Research On Stable Tracking Of Multiple Objects Technology In Complex Environments

Multiple object tracking,as an important task in computer vision,refers to inputting a video and tracking the trajectory of one or more types of objects without any prior knowledge of the object.Multiple object tracking has extremely important application value in intelligent surveillance,automatic driving,medical diagnosis and military visual guidance.This paper focuses on the multiple object tracking in complex environments,with a focus on tracking by detection algorithms.The stable tracking of multiple objects in complex environments is realized by researching current relevant tracking algorithms and improving the problems such as missed detection and false detection of the objects and low tracking accuracy.The main work contents of this paper are shown as follows:(1)An object detection algorithm based on YOLOv5 is proposed in this paper.The detection algorithm introduces the coordinate attention mechanism module into the YOLOv5 detection algorithm,which enriches the position information of the object.This algorithm can effectively solve the problems of missed detection and false detection of objects that often occur in complex environments,and has high detection accuracy,providing strong support for subsequent tracking tasks.(2)A multiple object tracking algorithm based on Deep SORT is proposed in this paper.The tracking algorithm uses the detection algorithm based on YOLOv5 to detect the objects,the Kernel Correlation Filter(KCF)and the Kalman Filter to predict the trajectory in parallel,the Mahalanobis distance and the cosine distance to match,and the successfully matched trajectory returns to the Kalman Filter update,finally forms the prediction-tracking-update system.The algorithm can effectively solve the problem of missing objects due to occlusion in complex environments,and has high tracking accuracy and close to real-time tracking speed in complex environments.(3)A stable pedestrian tracking algorithm based on Byte Track is proposed in this paper.The tracking algorithm uses the detection algorithm based on YOLOv5 to detect the objects,and the ORB algorithm to extract the features.This algorithm can solve the problems that the lack of feature information of small objects and the limited precision of feature extraction algorithms.It can stably perform multiple object tracking in complex environments and has high tracking performance.It has been verified that the accuracy of the detecting and tracking in this paper has been improved,and stable tracking of multiple objects can be achieved in complex environments.The algorithms proposed in this paper have excellent tracking performance in ablation experiments and comparative experiments on MOT16,MOT17,and MOT20 datasets,and the visualization of tracking demonstrates the robustness of the tracking algorithm application in practical scenarios.