Multi-target tracking method and system based on deep learning

Abstract

The invention relates to a multi-target tracking method and system based on deep learning, and the method comprises the steps: enabling a spatial pyramid module to be embedded into a multi-scale feature pyramid network, and obtaining an improved feature extraction network; acquiring images at T moments in a target scene and inputting the images into the improved feature extraction network to obtain target position detection results and target feature vectors at the T moments; calculating the intersection-to-union ratio based on the target position detection result and the prediction result at the t moment; calculating target feature cosine similarity according to the target feature vectors of the two moments before and after; screening out a plurality of tracking target objects according to the intersection-to-union ratio and the cosine similarity, and endowing the same tracking target object with the same label; and carrying out adaptive weighting and updating on the target feature vector at the t + 1 moment until each tracked target object in all the images is endowed with a label, and carrying out multi-target tracking according to the labels. A spatial pyramid and a self-adaptive feature updating mechanism are introduced, and the accuracy of multi-target tracking is improved.

Description

Technical field [0001] The invention relates to the field of machine learning, in particular to a multi-target tracking method and system based on deep learning. technical background [0002] Multi-target tracking technology is one of the important branches of computer vision. Its main function is to traverse the obtained video images to find the positions of the interested independent moving targets with some obvious visual features, and track them in the next video frames. Therefore, multi-target tracking technology is widely used in various fields. Such as intelligent human-computer interaction, monitoring and security, unmanned driving, intelligent navigation and positioning of robots, cruise missile guidance and other important tasks. [0003] At present, traditional multi-target tracking methods and deep learning methods have higher accuracy and less identity transformation for most scenes. However, the accuracy and robustness of the existing multi-target tracking methods n...

AI Technical Summary

Problems solved by technology

However, when the scene is complex, the crowd is crowded, there are too many interfering objects, and the target faces deformation and illumination changes, the accuracy and robustness of the existing multi-target tracking methods still need to be further improved.

Therefore, for the problem of unsatisfactory multi-target tracking in complex scenes, brightness changes, noise interference, occlusion between targets, etc.

Images