Key Technology Of Subway Obstacle Recognition Based On Multi-sensor Fusion

As an important part of urban public transportation,the subway has greatly alleviated the increasingly serious urban congestion.With the continuous development of autonomous driving technology in recent years,the development and application of high-level unmanned driving systems of the subway have received extensive attention.Although the subway has relatively independent road rights,the driving environment is not absolutely safe.From the perspective of multi-sensor fusion,this paper discusses the obstacle detection problem in the low-light and closed environment of subway tunnels.The paper mainly adopts the method of fusion of thermal imager and optical camera.It is hoped that an appropriate image fusion method can be selected to preserve the morphological features of the object with temperature difference on the thermographic image while preserving the detailed features on the optical image,thereby increasing the readability of image and improving the accuracy of obstacle extraction in the subway tunnel environment.The main contributions of this paper are as follows;(1)Data registration between the thermal imager and the optical camera.The main tasks of data registration include registration of data time and space.In terms of time registration,this paper mainly adopts the method of timestamp registration.In the spatial registration,the joint spatial registration of the thermal imager and the optical camera image coordinate system is carried out by using the world coordinate system as the transitional coordinate system.In addition,a calibration method of thermal imager based on Zhang Zhengyou calibration method was proposed,and the special calibration plate suitable for thermal imager was fabricated.(2)Research on image fusion methods.In this paper,the fusion of thermal imaging images and optical images is carried out,and the common image fusion methods are studied.In addition,an image fusion method based on IHS and wavelet transform is proposed.In the quality evaluation of fusion results,this paper adopts the objective quality and subjective image quality evaluation method while considering the purpose of fusion.(3)A hardware platform suitable for the extraction of obstacles in the subway environment.The selection of hardware and the construction of the fusion platform are carried out.In the construction of the hardware fusion platform,the scalability of the later sensors is considered.(4)Adaptive search of the starting point of the subway track.According to the morphological characteristics of the subway track,this paper proposes an adaptive tracking algorithm based on the integral of the track start point,which can effectively improve the accuracy of the limit extraction.(5)Preliminary extraction of obstacles in the low-light and closed environment of the subway.Under the condition that the conventional obstacle extraction method is not effective,the low frequency component in the wavelet image decomposition method is introduced,and the morphological characteristics of the obstacle are strengthened.Improve the recognition rate of obstacles in the low-light and closed environment of the subway.