Summary: | 碩士 === 淡江大學 === 電機工程學系碩士班 === 98 === This thesis is to use the Texas Instruments TMS320C6713 digital signal processor, the vision module VM480CCD, and the related software systems (e.g., Code Composer Studio) to obtain the task of the penalty kick of a humanoid robot by using neural network based localization. In this thesis, there have four parts: the path planning of gait, the image processing, the modeling using neural network, and the strategy for visual navigation, to execute the task of the PK for an HR.
First, the CCD module will capture the visual image, which is transferred to the TMS320C6713 for the image processing, including binary segmentation to reduce the storage and computation load, median filter to remove noise, image restoration to improve the accuracy, and calculation of the target position. The modeling using neural network is applied to establish the relationship between the image plane coordinate and the world coordinate. When the robot reaches in the vicinity of target (i.e., about 10 cm), the visual system starts searching the gate and the virtue target point to modify the posture of the HR for the PK. After the posture revision, a fine visual window is employed to confirm the posture for the PK.
The most important thing for vision system is the accuracy of localization. Therefore, a neural-network-based active embedded vision system is developed to approximate the relation between the world coordinate and the image plane coordinate. An interface for man and machine is also applied to design the desired motion of the HR, to connect the signal between the embedded vision system TMS320C6713 and the central embedded system RB-100, and then to navigate the HR to the planned posture for the PK. Finally, the corresponding experiments for the PK of an HR confirm the effectiveness and efficiency of the proposed system.
|