| Summary: | 碩士 === 國立臺灣科技大學 === 機械工程系 === 107 === This research aims to build an effective system for indoor service robots to search for objects. The system is formed by an integrated easy-to-operation structure, which includes an object initial-registration module and real-time in-motion registration module when an object appears to inside the view of the robot.
After the robot builds the indoor map, the users provide the images of the objects to be searched for registration, the vision system will register the positions of the objects in the map, and the data set embedded to the map can be used by the robot for object searching. The object registration system uses RGBD camera to capture the image of the object and removes the background before the SIFT features of this object are created for recognition and the recognition bounding box is used for definition of the object position. In order to reduce the ill-effect of varied illumination conditions to the object recognition, this research implements DBSCAN technique to stabilize and reduce the error of the object positions. This innovative object registration and search system can be integrated to SLAM navigation systems, and provide the effective object search capabilities.
|
|---|
