Multi-target tracking method and system based on random channel adaptive attention mechanism

Abstract

The invention belongs to the field of computer vision, particularly relates to the field of video multi-target tracking, and particularly relates to a multi-target tracking method and system based on a random channel adaptive attention mechanism. The method aims at solving the problem that in the target moving process, a target is often partially or completely shielded by an obstacle, and large displacement exists in pedestrian positions before and after shielding, so that continuous and stable tracking cannot be achieved after reproduction. The method mainly comprises four modules: a random module, an extrusion module, an excitation module and a weighting module: firstly, a gating mechanism is added during feature transfer and feature aggregation to realize random starting of a spatial adaptive attention module, global average pooling compression processing is performed on a transferred feature map after starting, and the obtained feature map has a global receptive field; and finally, calculating the weight of each channel by using a full connection layer, and weighting the original feature map, thereby enhancing the effect of the salient features in the image, enhancing the anti-shielding and long-time tracking capabilities of the model, and finally realizing the requirement of rapidly and stably tracking a video target.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to the field of video multi-target tracking, and specifically relates to a multi-target tracking method and system based on a random channel adaptive attention mechanism. Background technique [0002] With the application and great success of deep learning technology in tasks related to visual multi-target tracking, the current framework of visual multi-target tracking technology based on deep learning is mostly a two-stage mode, that is, data association based on target detection results to achieve multiple DBT mode for target tracking. In addition, from the perspective of the structure of the deep neural network, the sub-modules in DBT, such as feature extraction, can be integrated into the target detection network. Based on the fusion of sub-modules in DBT, joint detection and tracking can be realized, that is, the JDT mode, which uses a deep network framework to realize vision. M...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a multi-target tracking method and system based on a random channel adaptive attention mechanism, and to establish an intelligent model that can quickly and accurately realize multi-target tracking in video surveillance, so as to solve the problem of the part that is often blocked by obstacles during the target movement process. Or completely occluded, there is a large displacement of the position of pedestrians before and after occlusion, which leads to the problem that continuous and stable tracking cannot be continued after reproduction

Images