3D Panoramic Stereo Vision Reconstruction

Stereo reconstruction and 360 vision are active research areas in computer vision. The 360 stereo panorama camera presented in this work is equipped with a pair of bio inspired dynamic vision line sensors rotating at high-speed (up to 10 revolutions per sec) to provide real-time 360 panoramic views. A single camera built up with this sensor has already proven to be able to scan 360 scenes with great accuracy several times per second, thanks to the efficiency of the event-driven sensing. As a further step the AIT Austrian Institute of Technology has developed an algorithm for the stereo reconstruction of the pair of scans taken with this sensor. This work presents the development of methods and tools for ground truth setup and the evaluation of the 360 bio-inspired stereo panorama scanner. Three different test cases were designed to create ground truth data, which can be used to assess the performance of the stereo camera. In the first test case the ground truth data was created by simulating the working principle of the stereo vision system on available 3D data from a laser scanner. In the second and third test cases the ground truth data from real indoor and outdoor scenes was created using a rotating laser scanner (Velodyne HDL-32E). Finally, the comparative methods and the evaluation results are presented.