Hierarchical tracking method based on increment supervised gradient descent

Abstract

The invention discloses a hierarchical tracking method based on increment supervised gradient descent in the field of computer vision. The hierarchical tracking method comprises the following steps: initializing; collecting a plurality of samples in Gaussian distribution near a target position marked by a first frame, extracting features, and training a hierarchical regression model; positioning atarget; in a new frame, collecting a plurality of samples in Gaussian distribution near the tracking result of the previous frame, respectively regressing the samples to the tracking result accordingto a hierarchical regression model, and fusing the results to obtain a positioning target; collecting training samples online; collecting samples in Gaussian distribution near a tracking result, andperforming hierarchical regression on the samples to obtain samples of other layers for regular model updating; updating an incremental model; updating the hierarchical regression model online in an incremental learning mode; carrying out iterative tracking; repeating the step 2 to the step 4 until the complete video tracking is completed to obtain a more stable tracking result which can be used in artificial intelligence equipment.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a visual object tracking method. Background technique [0002] In recent years, fields such as computer vision, artificial intelligence, and machine perception have developed rapidly. Visual object tracking, as a fundamental problem in computer vision, has also been greatly developed. Object tracking is the process of using computer algorithms to process and analyze the images of consecutive frames in the video, so as to automatically track the position of the object of interest in the video, so as to obtain the stable trajectory of the object within a period of time. Traditional algorithms build complex tracking algorithms by building complex motion models, appearance models, memory models, and feature expressions. In recent years, with the widespread use of deep learning, many simple and effective tracking algorithms based on deep learning have been continuously developed in rec...

AI Technical Summary

Problems solved by technology

When the target is occluded or disturbed by the background, this type of method is easy to get offset tracking results, and it is difficult to achieve stable and robust tracking results in real scenes

Images