On the Fly Plane Detection and Time Consistency for Indoor Building Wall Recognition Using a Tablet Equipped With a Depth Sensor

This paper presents an on the fly planar segmentation algorithm that runs on a tablet equipped with a depth sensor and which uses a motion tracking algorithm. Our algorithm segments each incoming point cloud from the depth sensor and it then updates a global model containing all of the previously identified planes. Consequently, identical planes are identified in successive frames. We use a fast segmentation algorithm that generates and merges smaller intermediate clusters to identify all of the planes contained in the incoming point cloud. We then give each plane a unique ID by computing an histogram of its parameters. These IDs are used as a key for storage in a hash map. Identifying similar planes in different frames will enable us to update the plane's borders and will serve to identify the walls of an indoor scene. Our algorithm enables us to perform a 3-D planar segmentation of a point cloud that is issued from a depth sensor in less than 200 ms. Moreover, we are able to estimate the maximal size of a room with a mean error inferior at 10%. This will serve as a basis to develop a 3-D reconstruction algorithm that can automatically generate, in real time a 3-D editable model of an existing building. The generated 3-D model will contain the principal structural elements (i.e., walls, doors, and windows) of the building. This algorithm has a a number of applications, from simple 3-D modeling to building energetic performance assessment.