Vehicle detection method based on deep learning and multi-layer feature fusion

Vehicle detection plays a very vital role in the environmental perception of smart driving car. In real driving scene, the existing methods have some problems in detecting occluded or small vehicle objects, and thus the detection accuracy cannot meet the actual needs. This paper proposed a new method to detect vehicle objects based on deep learning and multi-layer feature fusion. First, building a convolutional neural network (CNN) to extract the advanced features. Then several layers with different shades of the network are selected for detection and preset candidate boxes are also used, so that detection model can adapt to multi-scale objects. In addition, multi-layer feature fusion structure is designed to enhance semantic information. The contrast experimental results indicate that our method has achieved better detection accuracy on test set, and the detection speed has reached real-time level, which can meet the requirements of smart driving car in practical application.