Research On Application Of Optical Flow Method In Train Speed Measurement

Train speed is an important parameter of the train control system.Traditional train speed measurement methods have their own shortcomings.With the increasing application of informatization and intelligent technology in various industries,it is important to study a train speed measurement method that meets the needs of intelligent rail transit.Engineering application value.In this direction,in view of the shortcomings of traditional train speed measurement methods,this paper uses the optical flow method in image processing to explore a new method of train speed measurement.The main contents are as follows:(1)Starting from the relationship between the optical flow field and the sports field,the basic theory of the optical flow method is introduced,including the constraints and derivation process of the basic optical flow equation,and the realization principles of the classical sparse optical flow and dense optical flow algorithms and the limitations of the optical flow method are analyzed,And then introduced the basic principles and related knowledge of optical flow network.On this basis,combined with the camera imaging model,a train optical flow velocity measurement model was established.(2)A fast optical flow calculation method that satisfies the train’s real-time requirements is studied.First,the distortion parameters are obtained through camera calibration to eliminate image distortion,and several feature point detection algorithms are analyzed;then,for the defect that the Lucas-Kanade(LK)optical flow method cannot calculate the large displacement motion,the image pyramid is introduced,and the image pyramid is introduced.,Layer-by-layer refinement method to solve the optical flow of large displacement motion;Next,in order to improve the real-time performance of the algorithm,it is proposed to use GPU to parallelize the optical flow calculation process;Finally,in order to facilitate the experimental verification of the algorithm,design circuit simulation software and conduct speed measurement simulation The experiment results show that the improved optical flow algorithm based on image pyramid improves the accuracy of optical flow estimation under large displacement conditions,and the parallel acceleration based on GPU can effectively increase the execution speed of the algorithm and meet the real-time requirements of the train speed measurement system.(3)Aiming at the fact that the LK optical flow method is susceptible to interference from the external environment and cannot obtain dense optical flow,a neural network is proposed to predict optical flow.Inspired by the idea of multi-scale residual learning,and drawing on the structure of the classic optical flow estimation neural network Flow Net,the optical flow prediction neural network model Flow Net2.0-MSD containing three sub-networks is built.The first sub-network extracts two input images The characteristics of the image and its correlation are calculated.The obtained initial optical flow is input to the second sub-network for optical flow refinement.The third sub-network is used to predict small displacements,and finally the two are merged to obtain a smoother large displacement light flow.The three sub-networks are separately trained and then fine-tuned using the Flaying Chairs,Things3 D and Chars SDHom datasets.The trained model is subjected to a train speed measurement simulation experiment.Due to the limitation of processing time,the optical flow network has a small speed measurement range but high accuracy.(4)On the basis of the algorithm proposed in the paper,the train speed measurement software was written,and the actual line experiment was carried out using the camera support platform built in the laboratory to further verify the performance of the two train optical flow speed measurement methods proposed in the paper.The experimental results show that the improved LK optical flow method based on GPU parallel computing has higher real-time performance and larger speed measurement range,while the real-time performance of optical flow network is relatively low,but the speed measurement accuracy is higher,which provides a new reference scheme for the speed measurement methods of relatively high speed and low speed trains respectively.