Study On Railway Foreign Object Detection Method Based On Structured Light And Its Hardware Implementation

In recent years,the railway industry has developed rapidly,the mileage of railway operations and the number of passengers and freight have increased significantly.The safety of trains is the fundamental guarantee for railway transportation.With the continuous increase of train running speed,the problem of railway foreign body encroachment due to human factors or natural disasters has become very prominent.However,the methods of observation and manual inspection by train drivers can no longer meet the requirements of the actual situation.Therefore,there is an urgent need to set up a foreign object detection system with high detection accuracy,fast speed and high cost performance on key railway sections to be able to find foreign objects in time and accurately.Under this background,this paper has carried out the research of railway foreign object detection method based on structured light,the main work completed is as follows:(1)The overall design of a railway foreign object detection system based on structured light is constructed.On the basis of deducing and calculating the system model parameters,the selections of lasers,cameras and other equipment required by the system are completed,and a structured light image acquisition platform that can be used for railway foreign object detection is built;this acquisition platform platform is used to complete the collection of more than 1400 images in four foreign object scenes,and the datas are annotated.(2)A railway foreign object detection method based on deep learning is designed.The traditional method based on maximum inter-class variance threshold is difficult to meet the requirements in railway foreign object detection.To this end,the YOLO-V3 network in deep learning is selected for foreign object detection.Since the performance of the network for detecting small-size targets is not outstanding,the anchor clustering and channel attention mechanism are introduced.The experimental results show that this method can effectively improve foreign object detection effect.(3)A dedicated embedded platform simulation and transplantation of railway foreign body detection method based on YOLO-V3 network are completed.Based on the analysis of computing resource requirements,the Hi Silicon Hi3519AV100 platform is selected to implement code development on the hardware acceleration unit(Neural Network Inference Engine,NNIE).The experimental results verify the correctness of the NNIE development,and the hardware platform can basically reach foreign objects real-time detection.This article shows that the structured light image acquisition platform,foreign object detection method research,hardware platform development and implementation are integrated.The prototype system is built and preliminary tests are completed,and it is expected to play a role in railway traffic safety and road network protection.