Vehicle Parking Pressureline Detection Based On Improved Convolutional Neural Network

This thesis mainly studies the illegal behavior of detecting vehicle stopping line pressing in the complex and changeable environment.For the traditional vehicle detection method,target features need to be extracted manually.In practical application,due to the interference of the external complex environment,such as rain,snow,light,background shielding and other factors,the traditional model not only has poor generalization ability,but also is not ideal for the detection effect of target features.Because of the convolutional neural network model and it,s later optimized model have excellent effects on target detection,convolutional neural network becomes the preferred algorithm in target detection tasks.Therefore,it is very necessary for us to study the application of convolutional neural network in vehicle parking line pressure detection.The main contents of the study are as follows:1)The structure and principle of the convolutional neural network series are introduced,and the methods of target detection by CNN model,R-CNN model,Fast R-CNN model and Faster R-CNN model are studied and analyzed.The application of Faster R-CNN model in target detection is emphatically introduced.2)The improved Faster R-CNN model is proposed for the problem of target detection with different size and the problem of missed detection of remote vehicle in image target detection.The improvement is mainly in the feature extraction.Fusion method and multi-scale image training method in the training process.Experiments show that the improved Faster R-CNN model has greatly improved the detection accuracy of the target vehicle,and also speeded up the detection.For vehicles of smaller size in the image,the detection capability is effectively improved by adding multi-feature fusion and multi-scale training.3)Firstly,this thesis trains the improved Faster R-CNN model with the KITTI dataset,then collects data in real life to test it,furthermore,analyzes the improved Faster R-CNN model and the number of layers of feature fusion,and train the relationship between image scales.