Abstract
Stereo vision systems imitate the mechanisms of natural human vision to get threedimension information from captured scenes. It is applied in the control of autonomous robots (to build a map of the environment) and the control of production tools. Stereo systems are deployed to identify objects and can be used for determining close-range depth. However, the error in estimated depth grows quadratically with the real depth. To increase accuracy, we must adjust the focal length and the baseline. Most cameras’ focal lengths are flexible to suit different scenes. However, once the baseline is fixed, it is inconvenient to change. Therefore, finding a suitable baseline is vital when designing a stereo vision system. To find the relationship between baseline and depth accuracy, the accuracy estimation needs to be analyzed based on stereo vision model and error propagation. Meanwhile, a measurement experiment will be conducted to confirm the relationship. The image data used in this project are captured by two GoPro cameras. Camera calibrations, as well as image operations and calculations are conducted using MATLAB. The accuracy of estimation is analyzed, which shows a simple relationship between accuracy and baseline. Based on this relationship, it is easy to determine a suitable baseline for different applications.