Comparative Analysis of Pothole Depth Estimation Using LIDAR and Stereo Imaging with Polarized Lenses: A DOE Approach

For vehicles with automated control systems to operate safely and effectively, pothole depth must be accurately detected. Potholes filled with water can cause serious accidents since they are concealed hazards that can cause vehicles to hydroplane. These risks are made worse by the fact that water-filled potholes can be misleading, particularly at night or in low visibility situations. This paper presents a comparative study of two methodologies for pothole depth detection: 1. Stereo Imaging using Cameras with Polarized Lenses. 2. Light Detection and Ranging (LIDAR) Technology. The performance, accuracy, and limitations of both approaches are thoroughly assessed under various surface and environmental circumstances using a structured Design of Experiment (DOE) methodology. In Challenging lighting situations, the stereo imaging method greatly improves depth estimate accuracy by utilizing polarization to increase contrast and decrease glare. This technique entails taking pictures from two marginally different perspectives, processing them, and producing a disparity map. By lowering reflections, the polarization aids in differentiating water surfaces and makes it easier to see how deep the pothole is the stereo imaging approach leverages polarization to enhance contrast and reduce glare, significantly improving depth estimation accuracy in challenging lighting conditions. Conversely, the LIDAR method generates high-resolution 3D point cloud data, enabling precise depth measurement even in low-visibility environments. The DOE framework meticulously defines key experimental variables, including surface texture, lighting, and measurement distance, ensuring a robust comparison between the two systems. Additionally, interaction analysis is conducted to examine how environmental and system variables collectively influence the accuracy and reliability of each method. Real time course planning and obstacle avoidance are made possible by the output data from both approaches, which are essential input for vehicle management algorithms.