Research On Key Technologies Of Intrusion Monitoring For High-Speed Rail Running Environment Safety

China has become the world's largest and fastest-operating high-speed railway country.By the end of 2021,the total operating mileage of my country's railways has exceeded 150,000 kilometers,of which the operating mileage of high-speed railways has exceeded 40,000 kilometers.The main characteristics of my country's railways are the large scale of the road network,covering different climatic and environmental conditions,and complex operation scenarios.In recent years,there have been many traffic safety accidents caused by the abnormal operating environment of high-speed rail in my country,resulting in heavy losses of people's lives and properties.Therefore,the safety and prevention of high-speed rail operation is extremely important.Through the analysis and conclusion of high-speed railway accidents,typical dangerous situations include perimeter intrusion,rockfall intrusion and natural disasters.A variety of monitoring systems have been deployed in the actual operation of high-speed railways to ensure the safety of the operating environment of high-speed railways.Among them,the monitoring system for natural disasters is widely used and has remarkable effects,but the intrusion monitoring for two typical scenarios of perimeter intrusion and rockfall intrusion is still insufficient.In recent years,with the rapid development of artificial intelligence technology and the iterative upgrading of environmental perception equipment,intelligence and precision are the development directions of high-speed rail operation safety monitoring technology.This paper starts from the typical high-speed rail intrusion scenarios—the practical problems of perimeter intrusion and rockfall intrusion,and conducts research on the key technologies of intrusion monitoring for the safety of high-speed rail operating environments.Facing the complex railway operating environment,analyze the applicability of different intrusion limit monitoring technologies,propose a high-speed railway intrusion limit monitoring method based on video and radar,integrate the detection results for decision-making,further improve the intrusion limit detection ability,ensure the safety of train operation,and reduce environmental safety risks.The main work of the paper is as follows:(1)Low illumination deep learning enhancement method based on illumination decomposition.Aiming at the problem that the target is difficult to detect in the surveillance video of the low-light environment,a multi-scale illumination decomposition and enhancement network is proposed by combining the illumination decomposition theory and the deep learning method to enhance the low-illumination image.First,the low-illumination image is decomposed into a reflection map and an illumination map by constructing a sub-network of illumination decomposition.Secondly,two image enhancement sub-networks are designed,and the illumination map and reflection map are respectively input into the enhancement sub-network for enhancement.Finally,the enhanced illumination image and the reflection image are synthesized to obtain an enhanced illumination image.Aiming at the problems such as artifacts and blurring that are prone to occur in the process of image enhancement,a multi-scale feature extraction module is designed to improve the network's ability to learn features of different sizes.The experimental results show that this method can effectively improve the brightness,contrast and details of low-illumination images.In terms of subjective and objective evaluation indicators,it has achieved the best results compared with other mainstream methods.(2)Video-based deep learning detection method for railway perimeter intrusion.In the railway monitoring video,the imaging area of different people from far and near is very different,and the complex and changeable natural environment will cause interference,which makes it difficult to detect targets in the perimeter.Therefore,an improved Fair MOT framework perimeter intrusion detection model is proposed.In order to improve the detection performance of pedestrians with different imaging sizes,the network receptive field is enriched by adding a receptive field module,and features of different scales are extracted;in order to improve the detection performance of pedestrians at night,the spatial attention module is added to fine-tune the features and extract more features.Characteristics of foreground pedestrians;improve the generalization and robustness of the method for railway scenes by training the network using a mixture of railway perimeter intrusion real data and pedestrian tracking datasets.Various experiments are carried out on the real scene data of railway perimeter intrusion,and the analysis results show that the improved method achieves the highest F1 score in the detection of targets of different sizes during the day or night,indicating that the perimeter intrusion detection method can be more effective.Applied to actual scenarios,it has practical value.At the same time,the joint detection with the low-light enhancement method can further enhance the perimeter intrusion detection ability of the detection algorithm in low-light scenes.(3)Intelligent detection method of railway perimeter intrusion based on millimeter-wave radar.Aiming at the problem that the video detection method has weak anti-jamming ability in special complex natural environment,based on the millimeter-wave radar with strong anti-jamming ability,an intelligent detection method of perimeter intrusion is proposed.First of all,in view of the high dimensionality of millimeter-wave radar data,which consumes a lot of computing power of the algorithm,the principal component analysis method is used to reduce the dimension of high-dimensional data.Secondly,aiming at the problem of optimal parameter selection of support vector machine detection model,a parameter optimization method based on reverse elite learning strategy combined with gray wolf optimization algorithm is proposed,and the optimal parameter value of support vector machine is found by this method.Experiments show that the detection model constructed by this method has a higher recognition rate and the fastest convergence speed.At the same time,the joint detection of perimeter intrusion is carried out based on video and millimeter-wave radar.The results show that the decision fusion detection in two ways has better recognition effect.(4)Video-based deep learning method for railway rockfall intrusion detection.Aiming at the difficulties of detecting foreign objects similar to the railway background and small-scale rockfall intrusion limit detection,a deep learning-based railway rockfall intrusion limit detection method is proposed.Aiming at the problem that the rockfall and the background are similar,the hybrid attention module is integrated to strengthen the network's ability to learn the features of the foreground target.For small target rockfall,the bidirectional feature pyramid module is integrated into the network structure,which strengthens the mutual communication between features at different levels.Simultaneously collect a large number of simulated rockfall data from different scenarios,build a simulated rockfall data set,and use the Mosaic data enhancement method in training to enhance the generalization ability of the detection method.The experimental results show that compared with various mainstream target detection methods,the method in this paper has achieved the highest accuracy,and the recognition results of different sizes of targets are stable.At the same time,joint detection with the low-illumination enhancement method can further enhance the detection algorithm's ability to detect rockfall intrusion in low-illumination scenes.(5)Intelligent method of railway rockfall intrusion detection based on lidar.Aiming at the problem that the video detection method has weak anti-interference ability in special complex natural environment,an intelligent detection method of rockfall intrusion is proposed based on the laser radar with strong anti-interference ability.Firstly,according to the large amount of invalid background point cloud data of railway scene,the point cloud data within the railway boundary is extracted through the method of straight-through and statistical filtering.Secondly,a random sampling consistency algorithm based on sample constraints is constructed to segment the point cloud of the orbital surface within the bounds.Finally,a parameter-adaptive clustering method is proposed to cluster the target point clouds of foreign objects such as rockfalls on the track surface,and obtain the results of foreign objects intrusion such as rockfalls.The experimental results show that the method in this paper can effectively reduce the calculation amount of point cloud data,and compared with other methods,the final detection results are obtained.At the same time,the joint detection of rockfall intrusion is carried out based on video and lidar.The results show that the decision fusion detection in two ways has better recognition effect.