EIoU: An Improved Vehicle Detection Algorithm Based on VehicleNet Neural Network
Zuomin Yang, Xianlun Wang, Jianguang Li
Abstract
Abstract The paper’s primary purpose is to optimize the performance (speed/accuracy) of vehicle detection. The vehicle dataset Vehicle2020 used in this paper is divided into ten different vehicle classes. Intersection over Union ( IoU ) is usually used as a standard to evaluate the accuracy of vehicle detection in a specific dataset. However, IoU as a performance of the object detection algorithm is still shortcomings. IoU is further improved and called a new algorithm EIoU . Finally, the neural network structure was redesigned, which was called VehicleNet. The experimental results show that EIoU as a performance evaluation algorithm used the vehicle detection framework can improve the performance of vehicle detection. Using the algorithm of this paper shows the performance superiority of vehicle detection.