A Multi-aspect Method for Vehicle Dynamic Detection Based On Deep Learning

In order to address the problems of dynamic target detection rate is low due to excessive interference of background areas and fast moving speed of detected targets in complex scenes,this article proposes a multi-aspect method for vehicle dynamic detection based on deep learning.The traditional deep learning algorithm is improved by using convolutional neural network with a multiplayer perceptron (MLP-CNN).The kernel of this improved method is first to apply the fast candidate region extraction algorithm to find the regions where vehicles may exist,then to utilize a deep convolutional neural network (CNN) to extract features of candidate region,and to use an enhanced convolutional layer with multilayer perceptron (MLP) to further abstract optimal features for each layer.The Support vector machine (SVM) is finally used to classify CNN features of backgrounds.The results show that the proposed method can deal with part occlusion or fast motion objects.With a recognition accuracy of 87.9% and elapsed time of 225 ms,it is more efficient than other traditional methods.