An object detection system based on YOLO in traffic scene

Abstract

We build an object detection system for images in traffic scene. It is fast, accurate and robust. Traditional object detectors first generate proposals. After that the features are extracted. Then a classifier on these proposals is executed. But the speed is slow and the accuracy is not satisfying. YOLO an excellent object detection approach based on deep learning presents a single convolutional neural network for location and classification. All the fully-connected layers of YOLO's network are replaced with an average pool layer for the purpose of reproducing a new network. The loss function is optimized after the proportion of bounding coordinates error is increased. A new object detection method, OYOLO (Optimized YOLO), is produced, which is 1.18 times faster than YOLO, while outperforming other region-based approaches like R-CNN in accuracy. To improve accuracy further, we add the combination of OYOLO and R-FCN to our system. For challenging images in nights, pre-processing is presented using the histogram equalization approach. We have got more than 6% improvement in mAP on our testing set.