Street View Vehicle Based On Mask R-CNN Research And Application Of Target Detection Methods

In the field of computer vision,target detection is a very challenging research topic.Vehicle detection is a branch of target detection and is widely used in road supervision and unmanned driving.Vehicle detection combines the positioning and classification of vehicles,using machine learning and image processing techniques to determine the target area of interest.Conventional vehicle detection performs feature extraction by direction gradient histogram and scale invariant feature transformation,and inputs the extracted features to the classifier.This method of manually extracting features as an image representation is difficult to adapt to changes in complex scenes,and the generalization ability is poor.In recent years,with the development of deep learning,the target detection algorithm based on convolutional neural network has gradually become the mainstream.Its advantage is that it can automatically extract features according to the data set,and it has a certain degree of invariance to changes such as deformation and illumination.The application of deep learning has improved the accuracy and speed of vehicle detection.However,there are still insufficient deficiencies in vehicle detection in scenarios such as small targets,occlusion,and shadow coverage.Based on the above analysis,based on the application scenario of actual vehicle detection,this paper studies the vehicle detection algorithm based on deep convolutional neural network,improves it,and preprocesses the data set to improve the detection accuracy,main tasks as follows:(1)For the actual street scene of this thesis,in order to obtain the most suitable target detection algorithm,the most representative target detection algorithm rmodels based on convolutional neural networks are studied.R-CNN,SSPNet,Fast R-CNN,Faster R-CNN,Mask R-CNN,etc.,and compared their performance with YOLO and SSD,and finally chose to use the Mask R-CNN with better effect in the actual scene with more small targets as the vehicle of this thesis.Detection algorithm model.(2)On the basis of the research on the principle of Mask R-CNN algorithm,the structure is improved:the network structure of feature extraction is designed,the efficiency of the algorithm is improved by reducing the number of network layers;the structure of candidate window classifier is designed.The use of bilinear interpolation reduces the error in feature extraction of the region of interest and improves the accuracy of vehicle detection.At the same time,the training data set is preprocessed-histogram equalization and image sharpening,so that the accuracy of small target detection in the scene is improved.(3)After preprocessing the 250,000 image datasets with the above method,the improved algorithm was trained and tested for actual effects,and compared with the YOLO test results.The test results show that the improved Mask R-CNN achieves 91.5%,90%,79.4%and 78%accuracy under different illumination,shadow,and different degrees and different parts of the scene,and the detection of small targets The accuracy rate has been greatly improved,and the accuracy rate of 78%has been achieved.The effect is obviously better than the YOLO algorithm,which has reached the research goal of this paper.