LiDAR Technology and AI Vision Technology for Railway Transportation Solution (Part 2-Technology)

In this article, we will discuss how LiDAR technology and AI vision technology work for tunnel entrance intrusion detection in railway transportation.

I. Technical Principle

1. AI vision technology

The system collects real-time images in front of the tunnel entrance through visual sensors and sends them to the system host. The host performs real-time processing, analysis, and storage on the source video image. The visual analysis module intelligently identifies foreign objects on the track and marks them in the video stream. The video stream is then transmitted to the data fusion processing module through an internal interface. This system is mainly used for detecting long-distance intrusions of foreign objects. The detection principle is shown in the figure below:

Visual Inspection Schematic Diagram

① Camera data collection module: to collect RTSP stream data information from industrial cameras through the LAN, and the system uses hard decoding to decode the encoded data;

② Target detection module: using deep neural network algorithm to intelligently detect obstacles in real time, the process includes feature extraction, border regression, and classifier classification;

③ Railway segmentation module: using deep neural network algorithms to intelligently identify rails in real time;

④ Prediction compensation module: using Kalman filter algorithm to track and compensate for the real-time position of obstacles;

⑤ Data fusion processing module: The visual obstacle detection results will be transmitted to the data fusion processing module. The detection scene is shown in the figure:

AI Vision Used to Detect Scenes

2. LiDAR Technology

Based on the high-precision laser echo signal measurement technology, the point cloud data in front of the tunnel entrance is collected to achieve three-dimensional imaging through data modeling. The newly obtained point cloud data is sent to the system host. The LiDAR processor combines algorithms to calculate and analyze (by comparing with existing data), and store the point cloud data detected by the LiDAR in front, obtaining information about the orientation, size, and distance of intruding objects. Based on its logical design, it provides early warning, mainly for detecting foreign objects at close range. LiDAR technology has technical characteristics such as high detection accuracy and accurate echo intensity.

II. Solution introduction

1. Solution introduction

Neuvition LiDAR is adopted in this solution, and the camera is the one specified by the customer or the conventional camera on the market matched with Neuvition Technology.

The project uses a high-resolution single-line LiDAR Titan M1 and a camera acquisition module to realize three-dimensional scanning and image acquisition of the tunnel entrance. The system obtains real-time point cloud data through LiDAR scanning, and then analyzes and processes this data, using clustering and comparison algorithms for real-time monitoring and early warning of intruding objects. At the same time, the system is also equipped with a video acquisition module for capturing real-time images and video data. The AI algorithm is used to identify intruding objects, and comprehensively process two recognition results to generate an early warning signal and trigger the on-site sound and light alarm module. In addition, the system has 5G wireless communication capabilities, which can transmit real-time point cloud data and image data to the cloud server for platform-level early warning.

At the tunnel entrance, we use LiDAR and a camera, supplemented by a host computer, to interact with the server through 5G wireless communication or optical fiber communication, to achieve comprehensive tunnel entrance safety monitoring and early warning services.

 Tunnel Entrance Intrusion Monitoring System Interface

2. System architecture

The LiDAR-based railway tunnel entrance intrusion monitoring system is divided into three parts: the sensing system, the processing system, and the result output system. The perception system uses high-resolution solid-state LiDAR technology and camera acquisition modules to collect the three-dimensional spatial information and video image information on site in real time, and then sends the point cloud data and image data to the processing system for processing. The processing system is implemented by an edge calculator, to perform real-time algorithm processing on the LiDAR point cloud data. If the preset safe distance is exceeded, the system will immediately trigger a real-time warning of on-site audible and visual alarms. At the same time, the real-time data on site is sent to the cloud platform through the 5G communication module, to realize remote viewing and coordination of on-site processing, and data can be saved to local or cloud servers. The on-site output results and the retrograde results are displayed on the display screen. At the same time, the display interface is equipped with an interactive control interface, which can set, control, and adjust relevant parameters of the system equipment.

System Architecture

3. Functions or advantages of the solution

• The LiDAR detection range can reach 200 meters, and the field of view is 45°*25°;
• Detection accuracy is 99%;
• The system outputs three-dimensional modeling data and image data of the tunnel entrance in real time;
• Output the size, speed, orientation, shape, and other information of the object;
• Linked sound and light alarm;
• Realize dynamic management and output data reporting of multiple tunnel entrances;