Object Pickup Manipulation Control of a Mobile Robotic System Based on Monocular Vision

This paper introduces a mobile robot system for object pickup in an indoor environment, including object recognition, target localization and navigation control algorithm models. A HSI based algorithm is presented here to realize object recognition, which makes use of Morphological filtering method to improve the stability due to multi-channel and multi-step segmentation. This algorithm is robust to varying illumination and image noises, and able to recognize accurately a target under very complicated environments. To localize the target in 3D space, an effective measuring algorithm is proposed in terms of the camera perspective geometric, incorporated with the horizontal plane constrains, and localization errors are reduced by a curve fitting method. Lastly, to improve the measurement accuracy further and circumvent the problem of blind area of vision, a semi-quantitive region division scheme is used to divide the field of interest into four measuring sections, which employs the auto-focusing function of the system camera. Experiments were conducted on a Pioneer mobile robot system with a PTZ camera, and the results show that the proposed schemes and algorithms are effective to complete a target object pickup task under varying and uncertain environments.