Abstract

RGB-D cameras have become important sensing devices for applications requiring simultaneous color and depth information, including three-dimensional reconstruction, human-computer interaction, robotics, augmented reality and computer vision. However, commercial RGB-D cameras use different optical sensing principles, leading to different mechanisms for estimating object depth. This mini-review examines the optics-based depth estimation principles of three Intel RealSense cameras: the D435, D435i, and L515. The D435 and D435i use active stereo vision, where depth is determined through epipolar geometry and triangulation using the disparity between corresponding image points captured by two infrared cameras. Their depth relationship is , where D is depth, f is focal length, B is the stereo baseline, and δ is disparity. Consequently, depth is inversely proportional to disparity. In contrast, L515 uses LiDAR-based time-of-flight sensing, which estimates distance from the propagation characteristics of emitted and reflected infrared light. For phase-based indirect ToF sensing, depth is calculated from the measured phase shift and modulation frequency. Numerical examples using camera parameters demonstrate depth calculations for the stereo and LiDAR approaches. By reviewing these two sensing mechanisms, this paper provides a concise theoretical foundation for understanding how commercial RGB-D cameras convert optical measurements into depth information.

Keywords:RGB-D camera; Depth estimation; Intel RealSense; D435; D435i; L515; Stereo vision; LiDAR