Multiple Object Tracking Algorithm Based on Adaptive Kalman Filter
TANG Chuanye
YAN Jiahao
CHEN Jianfeng
WU Haojie
Abstract：An improved multiple object tracking algorithm is proposed to solve the problems of inaccurate prediction of motion model,poor accuracy of state estimation and insufficient real-time performance in the process of multiple object tracking of road pe-destrians.In the camera motion solving,the LK optical flow method is introduced to combine with Fast feature points to accelerate the motion solving.The camera motion compensation is adaptively performed according to the prediction results of the Kalman filter based on the constant speed model to improve the accuracy of the motion target model.In the update stage of Kalman filter,the ob-servation noise is adaptively adjusted according to the confidence score of the detection results to adapt to the changing observation noise.The experimental results on the MOT16 dataset show that compared with the ByteTrack algorithm,the MOTA index is in-creased by 0.9%,the HOTA index is increased by 2.7%,the IDF1 index is increased by 3.8%,and the running speed is increased by about 47%,significantly improving the effect of multiple object tracking.
Keywords：multiple object trackingadaptive Kalman filtercamera motion compensationadaptive noise
Publication Date：2025-09-20
Online Publishing Date：2025-12-23(First online date of this platform, not the publication date of the document)
Pages：6( 2416-2421 )