Research On The Identification Algorithm Of Urban Rail Vehicle Number Based On Machine Vision

Identification of urban rail trains is a key step in the automated maintenance of trains.Most of the current train identification systems are based on RFID technology.These systems consist of a tag installed on the bottom of the train and a tag reading device installed on the ground.However,the label can easily fall off and get damaged.Besides,these systems also suffer disadvantages such as the high equipment cost and the complicated maintenance process.Therefore,it is of great significance to develop a train number recognition system with low cost,easy maintenance,high recognition efficiency and high accuracy.Based on the requirements of the train maintenance department for the train number recognition system,this paper introduces an image acquisition system with an industrial camera as the core,and investigates the train number picture positioning,segmentation and recognition algorithms to realize the automatic identification of the train number.Firstly,the overall design of the train number recognition system is determined,including the design goals,system composition,and working principle.The functional modules of the system hardware are studied,the key equipment is selected,the host computer acquisition camera-supporting system,as well as the corresponding train number recognition software are programmed.Secondly,the train number positioning algorithm is researched,and a train number positioning algorithm combining a maximum stable extreme value area algorithm(MSER)and a stroke width transformation(SWT)is proposed.This method uses MSER algorithm to extract the candidate area of train number characters,and uses heuristic rules and SWT algorithm to filter non-train number area.The remaining candidate areas are combined into train number text lines based on the nearest neighbor algorithm.The experimental results show that the accuracy of this approach reaches 96.56% in the complex environment such as uneven illumination and uncertain brightness.Then,the method of train segmentation and recognition is studied.A regional binarization method based on MSER is proposed for binarizing train numbers.This method solves the problem of the difficulty for the thresholding-based binarization methods to accurately complete binarization tasks in complex environments.The binarized train numbers are then segmented into individual characters.Experiments indicate that the segmentation accuracy reaches 97.86%.In the train number character recognition task,a pre-trained LeNet-5 based convolutional neural network(CNN)is used for automatic identification of the train numbers,with the accuracy as high as 99.92%.Finally,a field test is performed to evaluate this system.Results suggest that the system can accurately identify the train number,and the identification accuracy is 99.11%.The average recognition time is 0.98 s,which meets the needs of practical use.