Object Detection in unmanned vehicle with End-to End Edge-Enhanced GAN and Object Detector Network

Abstract

This paper proposes an efficient method for image detection for unmanned cars based on vision, and solves the problem of false localization for unmanned cars. The current SR methods based on deep learning have shown remarkable comparative advantages but remain unsatisfactory in recovering the high-frequency edge details of the images in noise-contaminated imaging conditions, we add Edge enhancement network (EEN) to GAN network to recover the high-frequency edge details. For the problem of false localization, we build a model of the bounding box of YOLOv3 with a Gaussian parameter and redesign the loss function. By using the predicted localization uncertainty and edge enhancement network, during the detection process, the proposed schemes can significantly reduce the FP and recover the high-frequency edge details. Compared to a conventional YOLOv3, the proposed algorithm, End-to-End Edge-Enhanced GAN and Object Detector Network improves the mean average precision by 4.2 on the COCO datasets.