An Implementation of Object Recognition Using Binocular Vision
The human active visual perception is closely related to a “fixation-move-fixation” rhythm. In recent past, the scientists adopted this concept with the expectation that camera-equipped robot can interact with the environment better by altering the viewpoint of cameras rather than passively observing the environment. In this thesis, a pair of pan-tilt cameras mounted on the ISAC humanoid robot was used to implement an active vision process along a one-dimensional search space. The software system including an object recognition sub-system was developed and was integrated with an arm control sub-system. The object recognition sub-system was designed based on color and shape detection and was implemented by using OpenCV libraries and MATLAB. Since two cameras were moving during the experiments, there was a linear mapping underlying the different combinations of angles of two sets of pan-tilt units for object localization. Several experiments were performed to evaluate system performance.