Research On Downstream Human Detection Of Expressway Tunnel Environment Based On Feature Extraction

The tunnel area is the key area of highway management,the pedestrian and non motorized vehicles entering the highway tunnel will seriously affect the work of the highway,resulting in huge security risks.According to the pedestrian detection technology under tunnel environment is an important guarantee for the work of highway.There are many poor conditions example for illumination conditions,a lot of noise in the image,pedestrian is the small goal,low pixel.What as above brings a great challenge to the detection for pedestrian in the tunnel environment.In this paper,we study the segmentation method of foreground and background in video and use the method of mathematical feature extraction and convolution neural network to detect the pedestrian target.According to the characteristics of the training classifier,the paper proposes a new search strategy which uses the motion information to reduce the search times.In addition,aiming at the difficulty of pedestrian feature extraction in the tunnel environment,this paper uses the advantage of the feature extraction of the convolution neural network to train the pedestrian detection network of end to end.The main contents of this paper as follows:(1)A single HOG feature is usually used in the training of pedestrian detection classifier,which often leads to low detection accuracy.This paper uses Local Binary Pattern model(LBP)characteristics and the Histogram Orientation Gradient features(HOG)serially input to train the classification model of support vector machine,through the hard cases of training steps,have combined features of training classifier based on greatly improving the accuracy of pedestrian detection.(2)The classifier based on HOG feature and LBP feature is generally used to search the whole image by sliding window,which results in great loss of time.In the highway tunnel,the monitoring picture appears in the fixed scene.According to the characteristics of video monitoring,by extracting information of pedestrian movement,the classifier detection and a modified Gauss mixed background subtraction method combining suspected motion region extraction in the image,the image classifier reduced search times,greatly enhance the efficiency of identification algorithm for pedestrian system.(3)In view of the highway tunnel environment noise caused by pedestrians and the environment contour is weak,the traditional machine learning method to extract effective features of the problem,this paper uses convolution neural network efficient feature extraction ability by improving the candidate frame extraction method,using the RPN candidate frame extraction network,single frame target recognition network training pedestrian detection under the condition of lacking in selection of single image candidate.The candidate frame extraction network and pedestrian detection network are trained to get the end to end pedestrian detection network.Compared with the pedestrian detection model,the accuracy of pedestrian detection is improved greatly,and the pedestrian detection speed is improved to a certain extent based on the RCNN algorithm.In order to meet the requirements of pedestrian detection in highway tunnel environment,this paper studies the adaptability of the classifier model based on feature extraction to the tunnel application scene,and puts forward the corresponding method to improve the detection method.This paper firstly applies the region convolution neural network to the pedestrian detection in the highway tunnel scene,and trains the end to end pedestrian detection model.It can improve the accuracy of pedestrian detection under the condition of tunnel monitoring,and it can be used as a reference for the identification of other objects.