Depth estimation from vision based wireless monocular camera sensor for soccer robot applications in MiroSot Middle League

Abstract

MiroSot is an area where robotics fields such as mobility structure, artificial intelligence. vision system, multi-agents, wireless communication and autonomous behaviour are of paramount importance. One of the most crucial parts of this overall system is the vision system. This system could determine the winning factor during a MiroSot competition. In this project, we are implementing a local vision system to determine the motion and behaviour of the soccer robot instead of a global vision system that is currently being used. Since there is limitation in the size of the mobile robot as stated in the MiroSot rules, hence some modification needs to be done on the vision system. A monocular vision system is used as an alternative to the more popular stereo vision system without neglecting the depth estimation aspect to determine ball distance and position. By using the CamShift algorithm to detect the ball, we use the ball diameter to estimate the distance of the ball. The calibration data has been tested and has shows that it gives accurate and precise distance estimation. By using this information, the soccer robot will try to avoid any obstacle by performing a sequence of movements. The details on how we obtained the calibration data for depth estimation and the testing phase have also been provided in this thesis. We also show how we test our control system and the results can be found in result chapter. The long term goal behind this project is to create a biologically inspired vision system for the soccer robot.