Research And Implementation Of 5G MEC System For Real-Time Object Tracking

Object tracking aims to accurately locate the position of the object in the video frame and track its motion trajectory.It is widely used in autonomous driving,abnormal event detection,augmented reality,etc.However,object tracking is time-sensitive.Whether the computing task is placed in the mobile terminal or remote cloud,the end-to-end delay cannot meet its real-time requirements.The integration of 5G networks and mobile edge computing(MEC)technology is an effective solution to ensure real-time requirements of object tracking.However,there are many problems in the real-world implementation of this technology.In addition,the 5G air interface transmission between the mobile terminals and the base stations is still the main performance bottleneck of object tracking.To this end,we propose two algorithms to eliminate the end-to-end delay and implement a 5G MEC system for real-time object tracking.The contributions of this thesis can be summarized as follows:(1)We design a transmission-inference parallelization algorithm based on pixel-neighborhood dependence.The algorithm models the end-to-end delay of object tracking and parallelizes the image frame transmission and inference process.The experimental results show that the algorithm significantly reduces the end-to-end delay of object tracking tasks.(2)We design a buffer-based resolution adjustment algorithm.The algorithm dynamically adjusts the frame resolution following the network fluctuations according to the sending buffer.When the buffer is large,the algorithm reduces the amount of transmission data volume by reducing the image frame resolution,thereby weakening the influence of signal fluctuations on a transmission delay.The experimental results show that the algorithm significantly reduces the transmission delay of video frames.(3)We implement a 5G mobile edge computing integration system for real-time object tracking.The system architecture is compatible with the latest ETSI standards and operational testbeds.Furthermore,we implement the above two algorithms in this system.The experimental results show that,compared with the traditional 5G scheme,the system reduces the end-to-end delay by 45.2%,and improves the object tracking accuracy by 12.7%,which can achieve a better real-time object tracking effect.