基于改进YOLOv7的动态手势识别探究  

Dynamic Gesture Recognition Based on Improved YOLOv7

作　　者：王强 王帅 胡明欢 Wang Qiang;Wang Shuai;Hu Minghuan(School of Big Data and Artificial Intelligence,Ma’anshan College,Ma’anshang,243100,China)

机构地区：[1]马鞍山学院大数据与人工智能学院,安徽马鞍山243100

出　　处：《黑龙江科学》2025年第4期94-96,100,共4页Heilongjiang Science

基　　金：马鞍山学院省级大学生创新创业训练计划资助项目(S202313614016);马鞍山学院2022年度校级科研基金重点项目(QS2022006)。

摘　　要：针对光照环境下动态手势识别研究中存在的识别准确率低、定位精度差及神经网络模型臃肿等问题,结合基于辅助边框计算边框定位损失方法,设计了一种基于改进YOLOv7网络框架的低光照环境下动态手势识别系统,将基于辅助边框计算边框定位损失方法与YOLOv7网络框架进行融合,改善了YOLOv7在低光照情况下识别精确率低和定位不准确的问题。研究结果表明,使用该系统能够完成在低光照环境下动态手势识别和定位跟踪任务,手势识别检测平均精确率达到99.4%,手势跟踪平均回报率在87.7%以上,综合任务检测跟踪成功率在93%以上,具有在光照下手势检测准确率高、定位准的特性。In order to solve the problems of low recognition accuracy,poor positioning accuracy and bloated neural network model in the research of dynamic gesture recognition under lighting environment,through combining with the method of calculating the frame positioning loss based on the auxiliary frame,low-light environment based on the improved YOLOv7 network framework is designed,and the problems of low recognition accuracy and inaccurate positioning of YOLOv7 in low light conditions are improved.Research results show that using this system can complete the tasks of dynamic gesture recognition and positioning tracking in low-light environments.The average accuracy of gesture recognition detection reaches 99.4%,and the average return rate of gesture tracking is above 87.7%.The comprehensive task detection and tracking success rate is over 93%,and it has the characteristics of high gesture detection accuracy and accurate positioning under light.

关 键 词：改进YOLOv7目标检测 动态手势识别 辅助边框方法 InnerIOU 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]