We have designed and implemented an accurate and robust obstacle detection system for use in the environment of an underpass tunnel. The system plans the collision area in front of the train through real-time location and mapping. The system is based on the IMU centered state estimator, which closely combines light detection and ranging (LIDAR), vision and inertial information with loosely coupled methods. The framework consists of three sub modules. Inertial measurement unit is regarded as the main sensor, which realizes the observation of LIDAR-inertial odometry and visual-inertial odometry to restrict the deviation of accelerometer and gyroscope. Compared with the previous pure point method, our method uses more geometric information by introducing line and plane features into motion estimation and using lines and points to utilize the environmental structure information.Object detection uses a combination of deep learning and point cloud clustering. Deep learning uses the projection of 3D information to 2D. As opposed to previous object detection, we put more emphasis on detecting the object and getting the spatial location of the object.Our system was experimented on a subway test line that is currently in use. The experimental results show that the system has some accuracy and stability in It may be able to play a crucial role in the safety of driverless subways in the future.
Subway Obstacle Detection System Based on Multi-sensor Data Fusion
Sae Technical Papers
SAE 2022 Intelligent and Connected Vehicles Symposium ; 2022
2022-12-22
Conference paper
English
Subway Obstacle Detection System Based on Multi-sensor Data Fusion
British Library Conference Proceedings | 2022
|Subway Obstacle Detection System Based on Machine Vision
British Library Conference Proceedings | 2022
|Subway Obstacle Detection System Based on Machine Vision
Springer Verlag | 2022
|NFLOW-Based Arbitrary Subway Obstacle Detection Method
IEEE | 2022
|