Layered multi-target tracking method based on significance detection

Abstract

The invention relates to a layered multi-target tracking method based on significance detection. The method comprises the following steps: constructing a mixed data set based on mixing of an existingstandard data set and manual annotation; constructing a road traffic scene salient region detection sub-network to generate a salient region and a non-salient region; detecting each frame of target byusing a target detection algorithm; constructing a convolutional network and a multi-target tracking network model combining long-term and short-term memory and a graph convolutional network, and tracking a salient region target; constructing a parallel KCF pool to perform single-target tracking on the target in the non-salient region; and combining the salient region with the non-salient regiontrack and performing post-processing to generate an overall track. A rapid significance detection method is added to generate a saliency region bounding box, and a target is detected and tracked through input; the detection precision can be maintained while the detection speed is increased; calculation complexity can be reduced for automatic driving multi-target tracking in a real scene, and tracking is accelerated.

Description

technical field [0001] The invention belongs to the technical field of automatic driving environment perception, and in particular relates to a deep learning multi-target tracking method, in particular to a layered multi-target tracking method based on saliency detection. Background technique [0002] Multi-target tracking is an important research field in the environmental perception of autonomous driving. It involves computer vision, sensor theory, communication theory and traffic engineering in practical applications, and can provide basic data for decision-making control after automatic driving. The main requirement of this task is to accurately track pedestrians and vehicles in road scenes. [0003] The performance of multi-target tracking consists of two indicators: tracking accuracy and tracking speed. Based on factors such as complex background, high density and random movement of targets, frequent occlusions and other factors in the actual complex traffic environme...

AI Technical Summary

Problems solved by technology

[0007]The purpose of the present invention is to overcome the existing multi-target tracking algorithm that consumes large computing resources and low computing speed, and provides a layering based on saliency detection The multi-target tracking method can reduce the calculation cost and realize fast multi-target tracking under the premise of ensuring a certain accuracy

Images