Object Perception For The Internet Of Vehicles Using Edge Computing

As the development of Internet of Vehicles,the combination of intelligence and communication for autonomous driving becomes prevailing.However,due to the limited capability of vehicles,it is difficult to implement the intelligent algorithms such as object detection directly on the vehicle.Although some studies suggested to use edge computing,it still faces tremendous challenges.First of all,it is still not clear the role of edge computing in Internet of Vehicles.Secondly,the penetration of object perception devices for vehicles is still limited due to the technique and price reasons.In our thesis,we proposed an object detection framework using Multi-access Edge Computing(MEC)based deep learning framework for the Internet of Vehicles(IoV)network in which the MEC is not only used for the computation of detect objects but also for the obj ect capture and transform the objects from MEC to IoV.In this manner,the vehicle does not need to equipment any devices for object capture and detection.The implemented system consists of four main parts,object detection,feature detection,feature matching points,and estimate the transformation model.The objective of this study is to investigate how accurate object detection and object transformation can be accomplished when the object detection of Internet of Vehicles is facilitated by edge computing.We combine two well-known deep learning frameworks at MEC that warns the driver that the current dynamics in relation to the ahead vehicles are focused on the imminent collision.We study the model,formulation and structure.In this process,we had performed the object detection at both sides successfully and transform the objects from edge to vehicle for IoV with high accuracy based on the collaboration between IoV and MEC.Our deep learning based MEC architecture transforms the object detected with Yolo and reduces IoV processing and network bottleneck with rapid and dynamic performance.We apply homography to estimate the transformation,which is best for smooth transformation between two images having the same scene and we get the final and appropriate image at IoV,which is an object transformed image.The performance of our experiment shows that our techniques YOLOv3 and SURF are robust for object detection and at the same time object transformation,which achieves better precision and complete transformation for IoV considered to a satisfactory outcome.