Vehicle Detection Method Based On Deformable Part Model And Convolutional Neural Network

Vehicle detection algorithm is an important research topic in the field of intelligent transportation and plays an important role in safe driving assistance technology.Because the deformable part model has a higher accuracy and better detection efficiency,it has been widely applied to the vehicle detection field.Under the current situation,the problem of reducing the rate of missed detection of partially occluded vehicles and the problem of reducing the false detection rate of non-vehicle vehicles have been two major technical difficulties in vehicle detection technology based on machine vision.Vehicle detection algorithms still have much room for improvement.In this paper,the vehicle detection algorithm based on the deformable part model in the traditional image detection is deeply researched,and mainly aimed at reducing the missing detection rate and false detection rate.The specific research work is as follows:(1)Based on the in-depth analysis of the principle of the deformable part model,this paper creatively proposes a vehicle detection algorithm based on the dual-vehicle deformable part model.The algorithm adopts the method of image region-differentiated matching and fusion of matching results to reduce the missed detection of partially-occluded vehicles detected by multi-vehicle detection.This dual-vehicle deformable part model vehicle detection algorithm can fully reduce the probability of missed vehicles.The experimental results show that this algorithm is superior to the existing algorithms in the detection of vehicles that partially occluded vehicles,and can meet the practical requirements in the application of safe driving assistance technology.(2)In view of the problem that non-vehicles in vehicle detection are mistakenly detected as vehicles,on the basis of building dual-vehicle deformable parts model,this paper creatively proposes a vehicle detection algorithm based on dual-vehicle deformable parts model and convolutional neural network.The introduction of the advantages of deep learning and efficient extraction of features into the vision-based vehicle detection field can better solve problems such as the incomplete design of currently manually designed vehicles.The convolutional neural network used in this paper is based on deep learning.The convolutional neural network used in this paper is based on deep learning to expand the feature extraction step of the deformable part model.One-to-one mapping of deformable part model feature extraction layer and convolutional neural network convolution layer,reconstructed as a convolutional neural network,replace HOG features used in deformable part models withfeatures learned from convolutional neural networks,enables the deformable part model to obtain more comprehensive vehicle characteristics,thereby reducing the false detection rate of the vehicle.The experimental results show that this algorithm can further reduce the probability of vehicle misdetection in the vehicle detection and make the vehicle detection algorithm more perfect.This paper trains a vehicle detection model based on dual-vehicle deformable components and convolutional neural networks,and validates the improved algorithm on the unmanned car platform of Jiangsu University.The source code is VS2008 project file,the operating system is Win7 64 bit operating system,and the development tools are Microsoft visual studio 2008 and Matlab 2012 a.