基于深度学习的车辆检测研究-计算机技术专业论文.docxVIP

优秀毕业论文 精品参考文献资料 Abst陀 Abst陀ct Abstract Vehicle detection is a challenging problem in the field of computer vision，which is to recognize vehicle and locate the position of vehicle in the given images or videos． With the rapid development of vehicle，the parking problem has become serious． Vehicle detection is the basis and premise for parking management，vehicle tracking， license plate recognition and other research．However,there are many difficulties in practice，such as occlusion，lighting and morphological changes．Vehicle detection requires higher accuracy and real-time，SO need to design a robust vehicle detection algorithm． Deep learning has been used in many fields，such as image classification，object detection，speech recognition and SO on，and has achieved amazing results．Vehicle detection is one of the object detection problem，and the applications of parking space detection has been more and more important，the main contributions of this dissertation are as follows： Firstly,the object detection algorithm based on deep learning in the parking lot is proposed．Some layers of convolution neural network are insensitivity for the target perspective deformation．We propose the A1CNN algorithm，which is the model of vehicle detection in the parking lot．As for the problem of vehicle position，we put forward the method of maximum height map to obtain accurate vehicle angle information．Experiments show that the proposed algorithm Can resist noise and Can detect the target vehicle more accurately． Secondly，we propose vehicle detection based on EdgeCNN for the image without calibration．First we extract proposals witll edge box as the training samples for the convolution neural network，the samples all come from the original images． For the test process，the test samples are input into edge box to get the proposals． After that，the proposals are put into the deep learning network to discfiminant．Our experiments show that EdgeCNN has higher average accuracy and robust than other methods