The Research Of High-Speed Trains Detection Algorithm Based On Video

The object detection is an important subject in the field of computer vision.The main purpose of this thesis is to detect and locate the target from the static image or video.It adopts the technology of pattern recognition,image processing,artificial intelligence and automatic control.It is widely used in road traffic monitoring,intelligent robot,automatic driving and so on.Because of the high-speed trains run at a high speed and the environment is complex and changeable,there is not a universal mature detection method for detecting high-speed trains.In order to assist the driver to protect the safety of train running,in the condition of moving camera,the real-time detection method of high-speed train ahead target is studied.It is of great practical significance to ensure the safety of high-speed train.In order to achieve real-time detection of high-speed trains,three widely used target detection algorithms are analyzed primarily,and then focus on the Adaboost algorithm for target detection and classification,and apply this algorithm to the detection of high-speed.In order to solve the problem of high false detection rate in the target classification and detection of the traditional machine learning methods,on the basis of deep learning,this thesis designs and implements the detection algorithm of high-speed train based on Faster R-CNN.The algorithm uses the RPN(Region Proposal Network)network to extract candidate boxes and share the convolution layer with Fast R-CNN for end-to-end training,improving the speed of feature extraction and the accuracy of detection degree.In order to realize the requirement of real-time detection of high-speed rail train,the method of detecting high-speed rail train based on YOLO is designed.The method can return the target frame and the target category directly to the different coordinates of the image.The speed of detection is increased.In this thesis,the data set of high speed train detection which based on the PASCAL VOC 2007 database is built.Then we build network model and the train data set is trained to verify the feasibility of the proposed algorithm.The average accuracy rate of high-speed train detection method based on Faster R-CNN is 92.6%,and the average accuracy rate of high-speed train detection method based on YOLO is up to 84.8%.The detection rate is higher than the average accuracy of the detector trained by the Adaboost algorithm.The detection speed of the high speed train detection method based on YOLO is faster than the Faster R-CNN method.The detection speed achieves 150 frames per second.It can achieve real-time detection of high-speed trains.The experimental results show the feasibility and effectiveness of two proposed algorithms in high-speed train practical detection.Finally,design and realize the high-speed traindetection system.The detection based on YOLO high speed trains are applied to NVIDIA Jetson TK1 development kit to train the actual video detection,which can display the test results in real time,and transfers the test results to the alarm module,and the voice reminds the driver.