Pedestrian Detection Based On Deep Learning

With the continuous development of society,the development of artificial intelligence level is also in an increasing state.In the machine vision,pedestrian detection has always been as the main research content,but also in the current intelligent monitoring,the system often need to use a technology,in recent years,with the continuous development of artificial intelligence,pedestrians Detection technology both in theory and in practical applications which have made great achievements.In the statistical study which is mainly based on the depth of learning,while the depth of learning is also the main direction for pedestrian detection research.In this paper,the basic theory of pedestrian detection and statistical learning is studied,and the feature extraction and classification methods of pedestrian detection are described respectively.But also through the use of the current mainstream method to complete the test experiment,so as to Based on the depth of learning to further get pedestrian detection method,the method is mainly based on the depth of learning as the main idea,through the convolution in the neural network to increase the deformation processing and occlusion processing layer,the purpose of doing so in order to better video images The detection model is divided into four parts,namely:image preprocessing,the establishment of convolution and sub-sampling layer,deformation processing,occlusion processing and classification.(1)Image preprocessing.This part mainly transfers the received image from RGB space to HSV space,and performs the reduction process in the process of transmission.Then,the gradient image is calculated by one-dimensional gradient operator,and finally the latest image feature data.(2)Establish convolution and sub-sampling layer.This part of the detection is mainly after the image preprocessing after the new feature data convolution and sub-sampling operation,the process requires two layers of convolution and a layer of sampling,and then the spheroidization of the convolution,so that you can achieve The adjustment of the weights and offset values in the convolution is intensified.(3)Deformation treatment.The corresponding deformation features are generated by using the mixed deformation model to obtain the new part of the image by convolution processing,and then the corresponding parts of the convolution and the corresponding deformation features are combined to obtain the new mixing feature And take the maximization function to detect the fraction of the generated site.(4)Occlusion treatment,classification.The model of the concept applied to the depth of the network of faith,and then the deformation of the site after the detection score for processing,which can be very good on the pedestrian occlusion factor tosolve the model parameters are RBM way training,the overall method of the model Tag classification layer,the data is the use of logical regression function to deal with.Based on the depth of the pedestrian detection model,this paper uses the sample in the INRTA database to carry out the relevant training,and then use the California Institute of Technology and the ETH database to carry out the relevant simulation test,and the simulation test out The results show that compared with the existing test methods,the final experimental results show that the pedestrian detection method model in this paper compared with the current detection methods,both in the detection accuracy,or in the missed rate and In the detection time have improved,to a large extent improve the pedestrian detection efficiency.