Research On Real-time Target Detection System In Edge Computing Sence

At present,a large number of mobile devices have been equipped with cameras.To process the captured images,mobile devices need to have the ability of target detection.Target detection technology could help the mobile device understand the content in the picture,and it is also the foundation of AR and other applications development.Most of the mainstream high-performance target detection solutions are based on convolutional neural network technology.Using mobile devices to perform these tasks will face problems of insufficient computing power,long delays,and high energy consumption.Edge computing proposed a new solution to these problems.Compared with traditional cloud computing,edge computing moves the data center to edge of the network.Users could offload complex computing tasks to the edge cloud to complete high-performance,low latency computation.Based on edge computing technology,this paper designs a high-performance real-time target detection system.The main achievements are as follows:1.This paper achieved a compression algorithm for continuous frames in videos.Based on the context information of the frames,the algorithm selects the region of interest in the image,and uses the nearest interpolation to downsample the non-interest region to reduce the image size.Experimental results show that this algorithm can get a good compression ratio and retain its key features.2.This paper achieved an intelligent offloading discrimination algorithm.By calculating the difference between adjacent video frames,the algorithm determines whether the current frame is a key frame,and offloads key frames to the edge cloud.Experimental results show that the algorithm significantly reduces the amount of offloading data,and hardly affects the accuracy of the target detection result.3.This paper designed the framework of the target detection system in the edge computing scenario.The system uses the above two algorithms and combines target detection and target tracking technology.This paper used the virtual machine to simulate the system.Experimental results show that compared with the traditional system,this new system can significantly reduce response delay,improve detection accuracy and reduce the pressure of edge cloud.