Generalized label multi-Bernoulli video multi-target tracking method based on SSD detection

Abstract

The invention discloses a generalized label multi-Bernoulli video multi-target tracking method based on SSD detection, and belongs to the field of computer vision and image processing. According to the method, the appearance of a target is expressed by using a convolution feature which does not need offline learning and has good robustness, and generalized label multi-Bernoulli (GLMB) filtering isadopted to realize video multi-target tracking. During multi-target tracking, the problem that the target tracking result is inaccurate due to the uncertainty of an unknown new target is solved. An SSD detector is introduced into a GLMB filtering framework to carry out preliminary identification on an unknown new target. The detection result and the tracking result are fused to obtain the final tracking result by adopting a weight summation fusion method, and the target template is adaptively updated, so that the problem of tracking offset in a filtering algorithm is solved, the problems of missing detection and false detection in a detection technology are solved, and the precision of a multi-target tracking state is greatly improved.

Description

technical field [0001] The invention relates to a multiple Bernoulli video multi-target tracking method based on SSD detection and detection of generalized labels, and belongs to the fields of computer vision and image processing. Background technique [0002] Video object tracking can be defined as given an initial state of a tracked object in an initial frame, and obtaining the state of that object in real-time in subsequent video frames. However, due to the diversity of target movement, occlusion, illumination changes, target deformation and complex environment, the target tracking problem has always been a difficult problem in the field of computer vision. Compared with single-target tracking, video multi-target tracking still has problems such as close or intersecting targets, especially unknown new targets and target disappearance, which further increases the difficulty of tracking. [0003] For the above-mentioned multi-target tracking problem, in the early days, the...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem of inaccurate target tracking results, even missed detection and false detection due to the uncertainty of new targets and complex environment interference in multi-target tracking, the present invention provides a single shot multibox detector based on SSD (Single ShotMultiBox Detector, SSD) Generalized label multi-Bernoulli video multi-target tracking method, the method uses SSD detection technology to detect the multi-target state of the current frame, and calculates the distance between the detection result and the surviving target; through the nearest neighbor algorithm matching, select the unmatched detection target as The new target is approximated in the form of a label Bernoulli set, and brought into the generalized label multi-Bernoulli filter for iterative tracking; in the tracking process, the distance confidence of the detection result and the filtering result and the similarity with the tracking target are calculated, And through the way of weight summation, the detection result and the tracking result are fused to get the final target tracking result

Images