Nowadays, applying cameras is becoming more and more common; especially because these cameras are becoming cheaper and easier to use. Also, the demand for cameras and automation is increasing in multiple sectors. Due to the fact that cameras are becoming cheaper and more common, different suppliers of cameras came up with each their own characteristics, where some cameras are even able to capture both RGB and depth images by using infrared or stereo matching for example. Nevertheless, the implemented stereo matching algorithms differ in accuracy by comparing for example the StereoLabs ZED 2i and the OAK-D cameras. This accuracy may depend on the stereo matching algorithm applied. Therefore, it is interesting to derive the depth ourselves by using deep learning stereo matching algorithms.
The task of this project is to investigate how depth could be reconstructed based on two adjacent images collected by a stereo camera. For this investigation, several aspects are important, such as the depth accuracy and the calculation speed. Furthermore, aspects such as the maximum depth, advantages, and disadvantages per algorithm are important to consider too. At the end of the project, a report needs to be written which clearly explains the different algorithms with their challenges and the differences between the stereo cameras used in the project.
20% Theory, 10% data collection, 50% implementation, 20% writing report
Le Viet Duc, firstname.lastname@example.org