Low-frame-rate video target tracking method based on target proposal

Abstract

The invention discloses a low-frame-rate video target tracking method based on a target proposal, and the method comprises six steps of S1, S2, S3, S4, S5 and S6. An input image is tracked through related filtering, and the tracking confidence degree is judged through the comparison of the time sequence information of adjacent frames. When the confidence coefficient is lower than a certain threshold value, it is considered that tracking fails, a low-frame-rate video target tracking method is provided, a target tracking system is designed, global searching can be conducted through a detection algorithm, a low-frame-rate target can be successfully tracked, the low-frame-rate target tracking method provided by the invention can still obtain a good tracking result under various scenes such astarget scale change, local shielding and light ray change, whether tracking is lost or not is judged through forward and backward tracking, and tracking interruption caused by target form change is prevented by comparing the classification scores of alternative target and obtaining the alternative target with the maximum response through related filtering when it is determined that tracking fails.

Description

technical field [0001] The present invention relates to a method based on the technical field of computer vision object tracking, and in particular, relates to a method for tracking a low frame rate video object based on an object proposal. Background technique [0002] Because the video image is relatively large and the video processing speed is relatively slow, low frame rate videos often appear, such as 3 frames per second. Such videos pose new problems for the field of computer vision because they are shot at a rather slow pace. For example, in the field of target tracking, the traditional video is 30fps. Because the frame rate is very fast, the targets in two adjacent frames appear in similar positions. Therefore, the traditional method is to search in the vicinity of the target to find the final target. However, in low frame rate video, because the time interval between two adjacent frames is relatively large, the distance of the target moving is relatively large, or ...

AI Technical Summary

Problems solved by technology

However, these traditional methods have two main problems: first, they cannot find potential targets very well

Because when the picture is large, if you use the traditional particle filter method, you can only search the entire picture with a sliding window, which will cause the search target to be too large, and it is impossible to find a suitable target.

Second, it is impossible to accurately determine whether the target is lost

Images