安徽工程大学学报2025，Vol.40Issue(2)：32-38,47,8.
基于改进图卷积网络的交警指挥手势识别研究
Traffic Police Command Gesture Recognition Based on Improved Graph Convolutional Network
摘要
Abstract
Traffic police gesture recognition is of practical significance for autonomous driving and assisted driving systems.The current traffic police gesture recognition method based on Graph Convolutional Networks(GCN)utilizes deep feedforward networks to process all skeletons in actions.This necessitates extensive floating-point operations for handling individual samples,limiting the model's performance due to computational constraints,resulting in poor practical applicability.To address this issue,a Temporal Attention Module(TAM)is proposed to enhance the efficiency of the model in traffic police gesture recognition.The computational effort is further reduced by early fusion of joint and bone data at the network input layer and integration of TAM into the GCN topology.Experimental results on the Chinese traffic police gesture dataset show that,compared with the original algorithm,the recognition accuracy of the proposed method reaches 90.73%when the computational cost is reduced by 4 times.The improved algorithm not only solves the problem of computing power limitation,but also provides an effective solution for traffic police gesture recognition in actual scenarios.关键词
时空图卷积网络/中国交警手势识别/动作识别/时间注意力/姿态估计Key words
Spatio-Temporal Graph Convolutional Network/Chinese traffic police gesture recognition/action recognition/temporal attention/pose estimation分类
计算机与自动化引用本文复制引用
张庆,梁涛年,时培成..基于改进图卷积网络的交警指挥手势识别研究[J].安徽工程大学学报,2025,40(2):32-38,47,8.基金项目
安徽省自然科学基金项目(2208085MF173) （2208085MF173）
安徽省重点研发项目(202104a05020003) （202104a05020003）
长三角科技创新共同体联合攻关项目(2023CSJGG1600) （2023CSJGG1600）
