一种改进YOLOv3的红外人体检测嵌入式系统的设计

Design of an improved YOLOv3 infrared human body detection embedded system

针对照明系统中对人体检测的需求,提出了基于改进YOLOv3的嵌入式机器视觉系统实现人体检测与定位。通过红外摄像头采集红外人体图像制作数据集;对YOLOv3进行改进并训练得到轻量化的人体检测模型,通过网格图方法由人体矩形框坐标得到人体的空间坐标。改进的YOLOv3人体检测模型的准确率提升了3.26%,参数量减少了96.01%、计算量减少了91.04%和模型体积减少了96.00%,经过模型转换后部署在带有神经网络处理器的K210嵌入式平台上,实现准确的人体检测和定位。在实际照明环境中对系统进行测试,系统的检测速度达到每秒16帧并取得了好的检测效果。实验结果表明,本文提出的人体检测和定位方法可以在嵌入式平台上实现且具有好的性能。

In response to the demand for human detection in the lighting system,an embedded machine vision system based on improved YOLOv3 is proposed to realize human detection and positioning.An infrared human body image is collected by an infrared camera to make a data set;YOLOv3 is improved and trained to obtain a lightweight human body detection model,and the spatial coordinates of the human body are obtained from the rectangular frame coordinates of the human body through the grid map method.The accuracy of the improved YOLOv3 human body detection model has increased by 3.26%,the amount of parameters has been reduced by 96.01%,the amount of calculation has been reduced by 91.04%,and the volume of the model has been reduced by 96.00%.After model conversion,it is deployed on the K210 embedded platform with a neural network processor to achieve accurate human detection and positioning.The system was tested in the actual lighting environment,and the detection speed of the system reached 16 frames per second and achieved good detection results.The experimental results show that the human detection and positioning method proposed in this paper can be implemented on an embedded platform and has good performance.