摘要
针对交警姿态估计存在的特征提取困难、实时性差等问题,提出一种改进的OpenPose网络交警姿态估计方法。采用MobileNet作为主干网络进行交警姿态的特征提取,解决模型随层次加深导致网络退化的问题,减少网络的参数量,加速主干网络内部特征的计算。通过跳跃连接机制将模型并行结构改进为串并同行结构,实现网络内部参数共享,降低模型的复杂度,提高检测实时性。实验结果表明,改进模型在COCO数据集以及公开交警数据集上分别获得78.9%和74.9%的mAP,检测速度可达25帧/s,为交通警察姿态估计问题提供了一种鲁棒性强、实时性高的实际应用方法。
Aimed at the problems of difficult feature extraction and poor real-time performance in traffic police pose estimation,an improved traffic police pose estimation algorithm based on OpenPose network is proposed.MobileNet was used as the backbone network to extract the features of traffic police pose,which solved the problem of network degradation caused by the deepening of the model layers,reduced the number of network parameters and speeded up the calculation of internal features.The parallel structure of the model was improved to the series-parallel structure by the jump connection mechanism,which could realize the parameter sharing within the network,reduce the complexity of the model and improve the real-time performance of detection.The experimental results show that the improved model can obtain 78.9%and 74.9%MAP on COCO dataset and open traffic police dataset,respectively,and the detection speed can reach 25 FPS.It provides a robust and real-time method for traffic police pose estimation.
作者
伍锡如
陈麒
Wu Xiru;Chen Qi(School of Electronic Engineering and Automation,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China)
出处
《计算机应用与软件》
北大核心
2025年第1期90-95,101,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61863007)
广西自然科学基金项目(2020GXNSFDA238029)
广西研究生教育创新计划项目(YCSW2020159)
桂林电子科技大学研究生教育创新计划项目(2020YCXS103,2021YCXS122)。
作者简介
伍锡如,教授,主研领域:深度学习,复杂网络,机器人控制。;陈麒,硕士生。
