Design And Implementation Of A System Of Detecting Pedestrian Invasion Ahead Of The Urban Rail Train

With the rapid development of urban track,the safety in the running of the train has attracted more and more attention.Keeping working a lot of hours,the driver may make mistakes.Recently,computer-aided-driving becomes a hot topic.At present,the domestic researches on this problem focus on physical contact protection and fixed-point monitoring in some key lines with the help of sensor technology.However,physical contact protection is restricted to short-distance situation,and fixed-point monitoring is also difficult to work in the whole running line.To better solve the problem,this thesis introduces a system of pedestrian invasion ahead of the train based on vehicle-sensor.In the running of urban rail train,we can obtain the dangerous area,warning area and safe area of the environment ahead of the train,determine whether there are pedestrians in front of the train,and make multi-level warning according to the area where the pedestrian locates.There exist two main contributions in this thesis,one is locating bound of different areas in front of the train,and another is detecting pedestrian which satisfies accuracy and real-time requirements.For locating bounds,a method of constructing running lines and mapping tracks is proposed in this thesis.Firstly,according to the curve information and slope information of line design and broken chain information of construction,the whole central line of track can be obtained by the means of geometry.Secondly,according to the train position,the section of central line in front of the train can be picked up,and the spatial position and shape of the corresponding track of this part of line can be obtained.Finally,the image location of the track ahead of the train can be obtained by the means of coordinate transformation.Taking line four of Hangzhou as an example,this thesis shows a convincing result of the method,and contributes a new idea of locating bound.For detecting pedestrian,this thesis adopts a method which joints radar and camera.Firstly,we use millimeter wave radar detection technology to get the target position.Secondly,we extract the target area in the image.Finally,we detect pedestrians in this area using YOLO neural network algorithm.Our method can reach the detection speed of 25fps on GTX1080,and it can obviously improve the detection accuracy of the neural network algorithm.The method of detecting pedestrians can meet the requirements of the system.