A platform and algorithm for omnidirectional robot sensing

Abstract

Computer vision and robot technology have been developed very fast in recent years. Although there are lots of algorithms and platforms to guarantee robots a robust perception now, most of them only focus on limited field of view. For example, they only focus on limited view angle in one direction. Therefore, when the light condition is not ideal, perception obtained from such algorithm and platforms may fail due to noisy input. To solve this problem, a new circle-ring platform is designed in this project. This platform is equipped with 8 monocular cameras aiming at different directions. These cameras are evenly distributed on the circle-ring platform and provide complementary information for each other. Two adjacent monocular cameras are paired to form a binocular camera to detect depth information of the photographed object. In addition to using stereo cameras to measure the depth of objects, LIDAR is also used in this project. Stereo cameras can get dense but relatively inaccurate depth information, while LIDAR can get accurate but relatively sparse depth information. Therefore, stereo cameras and LIDAR can complement each other in depth measurement of objects. In order to verify the fusion between sensors (stereo camera) and LIDAR, a binocular camera and a LIDAR are used for experiment.