Semantic mapping for smart wheelchairs using RGB-D camera

The previous perception and control system of smart wheelchair barely implement object related navigation such as furniture docking or door passage with interference from the obstacle avoidance behavior. In this article, a local 3D semantic map is built online using a low-cost RGB-D camera, which can provide category, geometry and functionality information of the recognized objects to the shared control module. The control system can choose different behaviors according to user intention to implement object related navigation. The steps include user intention estimation, target selection, motion control, as well as share control parameters adjustment. A flexible and scalable semantic reasoning based on forward chaining is used for object recognition, which utilizes the relationship of different objects to improve the accuracy of recognition. The proposed method is implemented on a smart wheelchair prototype equipped with a low-cost Kinect, which provides 3D point cloud for semantic mapping and fake laser for obstacle avoidance. And this method is tested in real environment, and the experiments illustrate that the semantic mapping for smart wheelchairs can effectively improve the collaboration between the user and the smart wheelchair and enhance the smart wheelchair's mobility.