Target tracking optimization method based on tracking learning detection

Abstract

The invention discloses a target tracking optimization method based on tracking learning detection. The target tracking optimization method comprises the steps of: determining L frame grayscale imagevideo frame sequences, wherein a tracking target position in the first frame grayscale image video frame sequence is known, and tracked target positions in the rest L-1 frame grayscale image video frame sequences are unknown; performing initialization, wherein t belongs to a set {1, 2, ..., L}, an initial value of t is 1, and L is a positive integer greater than 1; determining a main body part ofa first frame tracking target and an edge part of the first frame tracking target; acquiring Nt tracking points from an edge part of a t-th frame tracking target and a main body part of the t-th frametracking target, and further acquiring a limiting frame tb<t+1> of a (t+1)-th frame tracking target in the (t+1)-th frame grayscale image video frame sequence; acquiring N<t+1><db> detection target limiting frames in the (t+1)-th frame grayscale image video frame sequence from the (t+1)-th frame grayscale image video frame sequence; further acquiring a determined position of the (t+1)-th frame tracking target; and adding 1 to the value of t until determined positions from the second frame tracking target to the L-th frame tracking target are obtained, and recording the determined positions astarget tracking optimization results based on tracking learning detection.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to a target tracking optimization method based on tracking learning detection, that is, a target tracking optimization method based on tracking-learning-detection (Tracking-Learning-Detection, TLD), which is suitable for video frame sequences Long-term tracking of single targets. Background technique [0002] In the field of computer vision, target detection and tracking has always been a challenging research problem, and is widely used in intelligent monitoring, visual navigation, human-computer interaction, military manufacturing and other fields; target tracking is simply understood as Under the conditions of the video sequence, the process of continuously discovering the moving state information such as the position and shape of the tracking target in the video sequence frame, the so-called tracking target is the person or object of interest in the video frame sequence, etc.; in d...

AI Technical Summary

Problems solved by technology

[0004] The traditional TLD target tracking algorithm consists of a detection phase, a tracking phase, and a learning phase. The final outputs of the tracking phase and the detection phase are the tracking target bounding box and the detection target bounding box respectively. The final output obtained from the comprehensive tracking phase and the detection phase is called The target bounding box; the overall performance of the traditional TLD target tracking algorithm is worthy of recognition, but the robustness of the tracking stage needs to be improved, especially the selection method of the tracking point directly affects Lucas-Kanade (Lucas-Kanade, LK) The tracking quality of the optical flow method; the traditional TLD target tracking algorithm uses a uniform grid to select tracking points. If the integrity of the tracking target is to be maintained, some interference background information will inevitably be introduced, which reduces the performance of the algorithm. Robustness; if a more representative corner detection algorithm is used to select tracking points, it cannot be well applied to some tracking scenes with simple outlines of tracking targets, sparse corners, and complex backgrounds, and it also reduces the performance of algorithm tracking. stage real-time

Images