Research On The Shunting Locomotive Assistant Operation Based On Machine Vision

In recent years,the rail transit field has been experiencing rapid development,while the shunting locomotive working scene is complex,forcing driver to undertake a large driving workload,existing many security problems.Based on the demand for shunting locomotive pedestrian intrusion detection and driving control strategy based on machine vision have been investigated in this work,the research contents of which are as follows:(1)Considering the abnormal conditions of shunting locomotive railway environment,image enhancement algorithms for foggy days,nights,and rainy days have been proposed.For foggy days,based on Retinex algorithm,the central filtering function has been modified to an improved bilateral filter,which can effectively reduce noise while retaining the edge,obtaining the log domain images.Then the images have been processed using the sigmoid function,which not only reduces calculations,but also ensures the authenticity of color.For nights,the light source area has been specifically treated by retaining in the HSV color space and reducing the light interference in surrounding area.After reversing color,the dark channel prior algorithm has been used to enhance the foggy-like images.The color bias phenomenon in the enhanced image can be improved by combining the Lab histogram information corresponding to the daytime image.For rainy days,the directional spectral energy curve has been obtained by using integer and Fractional Fourier Transforms,and the direction of raindrop has been extracted.Non-raindrop imprint can be filtered out from the rotated image.Enhanced rainy images can be obtained by using the filtering algorithm.Through the evaluation of the experimental of different environmental images,the effectiveness of the proposed algorithm has been verified.New algorithms can improve the image contrast,restore color information,and supply clear images for the driver.(2)The brightest point region growing algorithm for extracting the railway track area has been proposed,and the rail history trend curve has been combined to obtain the shunting locomotive pedestrian intrusion area.The relationship model between image pixel height and distance has been established.The image below the vanishing point has been divided into close and distant area.For close area,PPHT has been used to detect the straight rail.While for distant area,the straight and curved rails have been extracted using the brightest point region growing algorithm.The accuracy in distant areas can be improved by using historical directional curves.Images collected during actual operation have been used for verification.The proposed algorithm can detect the rail position more accurately,quickly and effectively.(3)This paper has proposed pedestrian intrusion detection algorithms for visible and infrared image.For visible image,considering the problem of remote detection in the case of missing information,an algorithm based on an improved CNN for detecting pedestrian intrusion has been proposed.The shallow edge features are combined with the gray image,which is then fed as the input of the CNN.The cross entropy is then combined with the learning rate to reduce the training process.For infrared image,an improved image saliency detection algorithm has been used to extract the significance region.The sliding window algorithm with centroid relocation has been used to quickly locate the highlighted region.Zernike moment has been used to judge the symmetry of the image and its similarities with pedestrian.The trained CNN model has been used to generate the results.The visible and infrared images have been registered based on the locations of the vanishing points,rail lines and Harris points.The proposed algorithms have been validated based on pedestrian datasets,demonstrating that the proposed algorithms can effectively improve the accuracy.(4)A binocular vision based optimal loading control scheme with fractional-order PID following speed curve has been proposed to improve shunting locomotive operation efficiency.The characteristics of shunting locomotive low-constant speed loading system have been analyzed,and the optical flow method has been used to judge the loading state of shunting locomotive.The binocular vision reconstruction algorithm has been used to analyze the stacking distribution and obtain the velocity guidance curve.The locomotive motion model has been improved by referring to the traction braking characteristic.Fractional-order PID parameters have been set by genetic algorithm,and then the controller has been used to follow the locomotive speed.The new scheme relieves the locomotive driver of frequent gear control.The binocular vision leaves out the flat material operation in loading.Fractional-order PID controller is superior to the traditional algorithm with better control characteristic and stability.