Search

The 3D perception system functions as one of the core components for the safe operation of autonomous-rail rapid tram. Addressing the limitations of LiDAR systems used for autonomous-rail rapid tram in complex urban environments, such as challenges in perceiving distant objects, insensitivity to color, and potential failure, this paper proposes a pure-vision monocular 3D perception system for autonomous-rail rapid tram. Comprising data preprocessing, model training and model deployment, this vision-based scheme with a full coverage from data collection to on-board deployment enables the 3D perception of obstacles around autonomous-rail rapid tram, thereby improving the operational reliability in complex urban environments. The test results based on the autonomous-rail rapid transit dataset and Waymo open dataset show that the system can effectively perceive the complex road scenes in autonomous-rail rapid transit, achieving a final 3D average precision (AP) of 0.53, reasoning a single-frame of image about 56 ms, and meeting the real-time requirements of autonomous-rail rapid tram for the obstacle perception algorithm.

In the driverless transport and operation system of surface mine, the roadside perception system is used to assist driverless vehicles by providing road condition information. The driverless system currently applied in mine trucks realizes roadside perception based on the multi-sensor fusion technology, consisting of cameras, laser radars and millimeter wave radars. However, this system has several drawbacks, such as high system cost, a complicated structure and poor robustness. In this regard, this paper proposes a roadside perception approach based on an improved DeepSORT algorithm. This approach involves using cameras to acquire image data on vehicles and pedestrians in the mine, which are accurately identified by the YOLOv5s algorithm. Then, the improved DeepSORT algorithm tracks the identified objects in real-time, enabling statistical analysis to provide various functions, including vehicle traffic statistics, abnormal parking detection and pedestrian intrusion detection. The proposed approach was tested at the No. 8 intersection of Xiwan Surface Mine of Shaanxi Shenyan Coal Co., Ltd. The results show that using a single sensor approach can effectively achieve the recognition and tracking of vehicles and pedestrians at mine intersection, reduce the complexity of the roadside perception system and save costs compared to the roadside perception technology based on multi-sensor fusion.

In the intelligent driving system of transport vehicles in open-pit mines, the data fusion of camera and LiDAR plays a vital role in the detection, recognition and tracking of obstacles. Aiming at the problem of camera and LiDAR spatial coordinate unity, that is establishing the correspondence between 2D coordinate system of image and 3D coordinate system of point cloud, this paper proposes a camera and LiDAR calibration method based on chessboard calibration board. Firstly, Zhang Zhengyou Calibration Method is used to calibrate the camera internal parameters, and the internal parameter matrix and distortion coefficient are obtained.Then the camera and LiDAR respectively extract the corner coordinates of the calibration plate edge, and multiple groups of the three-dimensional coordinates of point cloud and two-dimensional coordinates of the image are obtained. Based on multiple groups of feature point pairs, the external parameter matrix is obtained by EPnP algorithm. In order to verify the calibration accuracy, this paper projects the point cloud on the image plane and compares its point pair accuracy. Experimental results show that, this method can obtain the position and pose relationship between cameras and LiDAR, and the projection error is less than 3 pixels, which solved the problem of camera and LiDAR space calibration of transport vehicles in open-pit mines.

Track component defect detection has long faced the problem of scarcity of defect samples. In the case of insufficient defect samples, the existing deep learning methods are prone to overfitting the model, and the generalization performance is poor, making it difficult to meet the actual track inspection requirements. To this end, this paper proposes a cascaded track component defect detection method which combines instance segmentation, fine-grained image classification and traditional image processing. Firstly, an improved fast instance segmentation network is proposed to locate fasteners and defective rails. Secondly, in view of the scarcity of defective fastener samples, a method of mask analysis and quantitative comparison of fastener contour lengths is proposed to realize the low-sample condition. At the same time, in view of the problem that the categories of rail defects are complex and subtle and easy to mix, a two-stage defect detection network of "instance segmentation + fine-grained image classification" is proposed, and label smoothing technology is introduced to improve model adaptability and classification accuracy. Experimental results show that, in the case of scarce samples, the defect detection method proposed in this paper has an accuracy rate of 95.7% for fastener breakage and loss detection, and 98% and 95% for rail defect detection rate and defect sub-classification, respectively. It meets the demand for track component defect detection.