A Fast Vehicle Detection Approach with Multilayer Perceptron Convolution Layers

Vehicle detection is attracting increasing attention. It is of great significance to propose a highly efficient and accurate vehicle detection algorithm. This paper proposes an improved method for vehicle detection based YOLO method. The new network adopts multilayer perceptron convolution layer to enhance nonlinear ability of feature mapping. In this new network, the fully connected layers are removed, and we predict the bounding boxes using anchor boxes. The new method effectively reduces model complexity and enhances object detection accuracy. Experimental results showed that the proposed method had 91.2% accuracy for vehicle detection under iteration number 20000 and had a great improvement than the other object detection methods.