基于改进NanoDet和GRU的交警手势识别研究  

Research on Traffic Police Gesture Recognition Based on Improved NanoDet and GRU

作　　者：于瓅[1] 马天祥 YU Li;MA Tianxiang(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《佳木斯大学学报（自然科学版）》2023年第5期10-14,共5页Journal of Jiamusi University：Natural Science Edition

基　　金：2021年安徽省重点研究与开发计划项目(202104d07020010)。

摘　　要：无人驾驶车辆对道路交通指示牌以及信号灯具有良好的识别效果,但在交警手势识别方面还有进一步发展的空间。当交警和交通信号灯同时在场以指挥复杂的路况时,能够识别出交警手势并根据交警手势优先做出判断尤为重要。考虑到车载设备的性能限制,使用了改进的NanoDet以保障实时从画面中提取人群中的交警区域,将注意力机制和GRU相结合,提高了该网络对交警手势的识别精度。设计了一种并行的运行架构,提高了交警手势智能识别系统的整体执行效率。Unmanned vehicles have good recognition performance for road traffic signs and signal lights,but there is room for further development in the recognition of traffic police gestures.It is particularly important to recognize traffic police gestures and prioritize making judgments based on them when both traffic police and traffic lights are present to command complex road conditions.Considering the performance limitations of onboard devices,an improved NanoDet is used to ensure real-time extraction of traffic police areas from the scene.The attention mechanism is combined with GRU to improve the recognition accuracy of traffic police gestures in the network.Design a parallel operating architecture to improve the overall execution efficiency of the traffic police gesture intelligent recognition system.

关 键 词：交警手势 NanoDet GRU 智能识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]