分散式救援機器人之即時影像防震與伺服定位

Abstract

由於救援機器人常行駛於崎嶇地形，造成機器人上影像晃動，因此發展了影像防震功能，使監控端可從機器人上收到一穩定的影像。在影像防震演算流程裡，加入Kalman filter估測自主運動向量，機器人即使在移動的過程中可得到一平穩防震的視訊；為了提升影像防震在崎嶇地形的穩健性，透過相關係數與變異量的分析，可明顯提升估測向量的精確度；此演算法已實現在TI DSP 6437，藉由階層式估測運動向量的機制，降低運算負載使防震效果更為即時，畫面更新率已由原來的10 fps提升至28 fps。 本研究也將合作搬運目標的功能實現在救援機器人上，使多機器人可自動趨近目標，並合作性地達成搬運的任務。透過多視角之影像偵測目標功能，設計融合機制降低感測器造成的誤差以建立精確的目標位置資訊，整合視覺伺服與虛擬邊界運動控制的機制，使多機器人精確到達目標點，藉合作機制可自動地完成搬運目標的任務。

Rescue robots integrating different sensors and the wireless network are applied to transmit sensing information for remote control. Among various sensors, the video renders the most important and useful information. When rescue robots navigate in a rough terrain, significant vibration of the video occurs unavoidably and a video stabilization system is proposed in this thesis to reduce the disturbances on the visual system of the present rescue robots. The Kalman filter is applied to estimate the motion vector of the robot and the false estimation occurrence can be thus greatly reduced by applying analysis of both correlation and variance of the motion vector estimation. With a hierarchical searching algorithm implemented on the TI DSP 6437, the frame rate can be improved from 10 fps to 28 fps in order to realize the real-time video stabilization. Furthermore, the vision techniques will be used to help multiple rescue robots accomplish cooperative mission. The cooperative mission of multiple rescue robots has been achieved by applying the developed visual servoing technique. For the target position estimation, triangulation is applied to construct 3D position. By the multiple vision systems, the fusion of the geometrical factors generates accurate target position estimation. The proposed virtual bounded motion control algorithm further leads the robots approaching the target precisely to move the target by the cooperation of multiple rescue robots.