Monocular Ranging Based on Camera Pose in Visual Train Positioning

Real-time and accurate positioning is the key to ensuring the safe operation of trains. Traditional train positioning technology relies on track-side equipment, and there are problems such as high construction and maintenance costs and difficulties in obtaining the initial train positions. With the rapid development of enabling technologies such as AI and image recognition, visual perception-based train autonomous positioning technology has attracted widespread attention in recent years. To address the problem of large measurement errors due to camera pose changes in train visual positioning, this article proposes a monocular ranging method based on visual beacon imaging size correction. A kilometer post is used as a visual beacon, and a deep neural network is used to detect the visual beacon and extract its feature points' pixel coordinates. The PnP algorithm is used to estimate the camera pose and correct the visual beacon imaging size through the camera pose angle, the average relative error of ranging is stabilized at 2% ~ 3%, and the ranging effect is improved by about 60%, which effectively reducing the ranging error of the monocular ranging model based on known references.