3D Object Recognition Using Texture and Color Features

• Main Authors: Kuan-Chia Chen, 陳冠嘉
• Other Authors: 陶金旭
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/32062839854104404364

碩士 === 國立中興大學 === 電機工程學系所 === 101 === In recent years, research areas on robotics have been flourishing and its application domain has been gradually extended from industry to home. In the trend of development of home robots, robots should be able to cope with the chores such as fetching objects and tasks related to housework. In order to perform these tasks, the robot should be capable of locating and recognizing the related objects. Since extracting key representative points and important features is crucial to an object recognition system, we propose a new descriptor for a regional characteristic of the object, gradient-enhanced fuzzy texture spectrum, and a hierarchical matching approach to increase the recognition rate and the matching speed. We will develop an active 3D object recognition system. The application scenario is to tidy up the room and put away the things which are placed on the floor. In the proposed vision system, a 3D image sensor is used for image acquisition. A new matching method, together with object distance measurement, image segmentation, will be implemented so that the robot can automatically detect and identify common objects in the house. By using the 3D image sensor, we can handle image segmentation more easily, so that the object matching can be more accurate. In addition, the robot’s mobility and the ability to move objects can be used to change the view angle or to move objects so that more information can be obtained for the decision with low confidence. Then better decision can be made with improved recognition accuracy.