基于边缘计算的实时目标检测系统的研究与实现

Research and Implementation of A Real-Time Object Detection System Based on Edge Computing

针对在小场景中灵活布设目标检测系统的需求,提出基于边缘计算的实时目标检测系统,采用边缘计算平台和目标检测算法相结合的方式对视频进行智能分析。选择低功耗嵌入式图像处理系统Jetson Nano搭建边缘计算平台,实现视频数据的边缘侧智能分析。利用优化后的MobileNet-SSD模型及特征融合模块在MobileNet-SSD网络上添加上下文信息,提高了目标检测的精度。在Jetson Nano嵌入式板上测试,该模型的平均精度达到80.4%,检测速度32帧/s,快于标准视频速度(30帧/s)。实验结果表明,该网络在检测小尺度和密度较大的目标时具有较好的检测效果。

Aiming at the requirement of flexible object detection system in small scene,a real-time target detection system based on edge computing is proposed.The edge computing platform and object detection algorithm are used to analyze the video intelligently.In this paper,Jetson Nano,a low-power embedded image processing system,was selected to build an edge computing platform to realize edge side intelligent analysis of video data.The MobileNet-SSD model is optimized,and the feature fusion module is used to add context information to MobileNet-SSD network,which improves the accuracy of object detection.Tested on Jetson Nano embedded boards,the model achieved an average accuracy of 80.4%and detected at 32 frames per second,faster than the standard video speed of 30 frames per second.Experimental results show that the proposed network has a good detection effect on small scale and large density targets.