An Improved Target Detection and Traffic Parameter Calculation Method Based on YOLO with a Monocular Camera

YOLO, which is one of the deep learning target detection methods, has been used in detecting traffic targets in real-time on regular devices under general light and medium traffic density. Compared with other deep learning methods like R-CNN, it was reported more efficient in computing, but less accurate in bounding box detection. In order to improve the detecting performance, this paper builds a hybrid method which combines frame difference (FD) method with YOLO. The result shows significant improvement in accuracy of bounding boxes, with little effect on detecting speed. This paper also proposes a model to calculate the speed and distance of targets in real-time based on a fixed monocular camera. The methods introduced in this paper can be used to warn drivers based on the traffic surveillance systems at intersections, or to help drivers avoid dangerous driving behaviors through on-board-devices.