| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 104 === This study proposes a real-time image processing technique, which is combined with robot kinematics and camera model. The proposed technique works on an eye-to-hand image processing architecture. The architecture can be used for the image objects at the robot end effectors. However, the wide field of view (FOV) in this architecture will cause low image processing speed, and color interference in images.
Therefore, to achieve the image registration and predict the position of the object in image, this study combines robot kinematics and camera model together. Due to the known 3D CAD model of the object, direct kinematics can calculate its three-dimensional position from the encoder feedback. In addition, the camera model can convert the 3D workspace to 2D image coordinate and must be calibrated using calibration board first. Then, this study can obtain a conversion relationship from robot coordinate to image coordinate. During the image recognition, the conversion relationship can calculate the object position in image through the kinematics information in real time and produce dynamic region of interest (ROI). In this way, this study can reduce the complexity of image processing and also reduce the color interference in image.
Moreover, a challenging stir frying robot is chosen to verify this system. This thesis applies the system to identify image features during the stir frying process. The results of this study can be used as visual servoing to automatically adjust parameters of stir frying robot in the future. For the experimental results, the study can verify the accuracy of pan connected to the end effector. Next, we can evaluate the area and the trajectory from center of mass for the spaghetti. At last, this study is verified to raise the speed of the image processing dramatically.
|
|---|
