Pedestrian target tracking method based on convolution association network in automatic driving scene

Abstract

The invention relates to a pedestrian target tracking method based on a convolutional association network in an automatic driving scene. Characteristics are shared by using a convolution association network and a lightweight target detection network, the relevance between the targets is captured, and then target tracking is achieved. The method specifically comprises the following steps: A1-1, obtaining a one-stage target detection model, performing down-sampling for five times in total; performing prediction on the feature maps of the last three scales; adopting a common convolution module except during first downsampling, replacing adopted models designed in a multi-scale downsampling module during subsequent four downsampling with separable convolution modules, performing target frame prediction on the models on a feature map of the last three downsampling, and finally forming a lightweight pedestrian target detection network is formed; A2-1, predicting a target incidence matrix toobtain a target incidence matrix; and A3-1, designing a pedestrian target tracking strategy through a convolutional association network, and finally obtaining a pedestrian associated target tracking result. The method can well adapt to an automatic driving scene in the aspects of tracking precision and speed.

Description

technical field [0001] The present invention relates to the technical field of multi-target tracking, in particular to a pedestrian target tracking method based on a convolutional association network in an automatic driving scene. Background technique [0002] With the increase of car ownership, the problem of road traffic safety has gradually become prominent. The automatic driving system can reduce the driver's operating load during driving, provide support and assistance for the driver, and improve travel efficiency; prompt and correct the driver's improper operation during driving, avoid traffic accidents, and save lives and property . As people's requirements for driving safety are getting higher and higher, autonomous driving technology has received common attention from academia and industry. In the autonomous driving scenario, it is not only necessary to detect the targets on the road, but also to track them and understand their movement trajectories, so as to cont...

AI Technical Summary

Problems solved by technology

In the automatic driving scene, the proportion of pedestrian targets in the image is usually very small, and the similarity between small pedestrian targets in the distance and roadside tree trunks is very high, which brings additional difficulties to pedestrian target detection. General target detection algorithm The detection effect on pedestrian targets is not ideal

Images