Traffic police gesture recognition method based on multi-view adaptive network of attention mechanism

Abstract

The invention discloses a traffic police gesture recognition method based on an attention mechanism multi-view self-adaptive network, comprising: step S1, collecting video data of traffic police gestures, extracting traffic police skeleton node data based on the video data, and establishing a skeleton node data set; step S2, constructing The multi-view adaptive sub-network adopts the multi-view adaptive sub-network to obtain the observation view coordinate system, and performs vector representation of the skeleton node data in the observation view coordinate system; step S3, based on the attention mechanism, the feature vector of the skeleton node It is constructed as a graph network data structure, and the features of the skeleton nodes in the graph network data structure are enhanced by using the extrusion and excitation SE modules; step S4, based on the enhanced skeleton nodes, the traffic police gesture is extracted using the spatio-temporal graph convolutional network ST‑GCN The spatiotemporal feature information of the traffic police gesture is used to recognize the traffic police gesture. The present invention can quickly and accurately identify traffic police gestures.

Description

technical field [0001] The invention relates to the field of unmanned driving technology, in particular to a traffic police gesture recognition method based on an attention mechanism multi-view adaptive network. Background technique [0002] Complex and changeable urban roads are one of the main application scenarios of driverless technology. At this stage, unmanned vehicle technology still needs to be continuously improved to adapt to non-ideal real-life scenarios such as bad weather and congested roads. According to the "Contents and Methods of Road Test Ability Evaluation of Autonomous Driving Vehicles", road test vehicles should have the ability to understand traffic command gestures. In other words, the unmanned vehicle must be able to correctly recognize the gestures of the traffic police in real time, and make real-time driving decisions corresponding to the gestures of the traffic police. [0003] At present, there are relatively few research methods in the field o...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a traffic police gesture recognition method based on the attention mechanism multi-view self-adaptive network, to solve the technical problems in the prior art, to detect the feature data of the traffic police skeleton nodes under the multi-view, and to use attention The force mechanism enhances the weight of effective node data, adopts an adaptive network hierarchical structure to fuse multi-view spatiotemporal feature data, improves the robustness of the traffic police gesture recognition method, and solves the problem that unmanned vehicles and traffic police cannot "communicate"

Images