Camera Calibration/Computer Vision Summer Research
This is my summer research project on stereo vision with Professor Daniel Scharstein for the Middlebury Stereo Vision benchmark project in Summer 2019
Skill: computer vision， c++， openCV，camera calibration
In 2019, I undertook summer research on stereo vision with Professor Daniel Scharstein for the Middlebury Stereo Vision benchmark project. Specifically, I was in charge of designing and optimizing the camera calibration step, which estimates the intrinsic and extrinsic parameters of the camera system using calibration boards, most commonly a chessboard, which is highly accurate, or an AruCo board, which is less reliable but allows for occlusion of the board. One particular challenge I faced was when I placed the board too far from the camera, and the board could not be properly detected. To solve the problem, I created, printed, and experimented with self-designed AruCo chessboards where I placed AruCo codes in the setting of a chessboard, and ChaRuCo boards, which combines the property of both a chessboard and an AruCo board, to increase detectability and accuracy. I also experimented with different settings and discovered that room setting, lighting, tilt of the board, the number of boards, the placement of boards, and the size of AruCo code all played a role in the detection of board markers. Finally, I wrote and refactored a program that prints the number of detected markers. The program prints out the number of detected markers while the user takes a set of images, output the number of detected markers, and warn the user to retake the image if the number of detected markers is too low. This specific step can improve the quality of data used for camera calibration, bringing more accurate result to the final analysis.
summer research project on stereo vision with Professor Daniel Scharstein
Github:
