Improvement of Door Recognition Algorithm using Lidar and RGB-D camera for Mobile Manipulator

The characteristics of mobile manipulator, which can perform various tasks in dynamic environments, have the advantage of driving and doing diverse tasks in large indoor environments with complex structures such as high-rise buildings. However, in order to efficiently navigate in such environments, mapping process containing information about diverse objects that robot can interact with is essential. Among these objects, door is of great importance, but door recognition is challenge because there are doors of various structures and sizes even in single indoor environment. This paper proposes an improvement of door recognition algorithm for mobile manipulator robot using RGB-D camera attached to end effector of manipulator and Lidar of mobile platform. Basically, laser scan data from Lidar is processed by line fitting algorithm and vision data from RGB-D camera is processed by YOLOv3. The process by laser scan data enables the first door recognition. And Additional recognition through vision data is possible by controlling the manipulator according to the weights given by the first recognition. The proposed algorithm has been verified in a simulation environment based on real-world, and we confirmed that it has a higher recognition success rate compared to traditional algorithms.