The Mobile Robot Indoor Scene Cognition Using 3D Laser Scanning And Monocular Vision

Mobile robot need to acquire environment information in real time during its movement, so how to implement effective understanding and cognition about the environment information is of great importance. Laser range finder is extensively used to acquire 3D information from environment for its high abilities of prevent interference and low influence of light. Additionally, camera has mature application in image acquisition. This paper takes laser range finder and camera as robot visual system and studies the indoor environment cognition problem for mobile robot.For the structure of indoor environment is structural, it can be represented by simple geometry feature. Line features can be extracted from the laser data by its storage order, and then neighbor lines are combined to get plane features. In our work, planes are extracted from 3D laser data using regional expansion algorithm, and the properties as well as the relationship of these planes are used for structure cognition. When the structure of the indoor scene is removed, the shapes of the objects in the remaining data are uncertain and rregular so it is hard to represent by simple geometry feature. We propose an object segment algorithm based on the continuity of the points, to implement effective segmentation of the objects in the indoor scene.Markov Random Filed theory is widely used in image processing and computer vision. According to the relations of the neighbored structures and objects of the scene, this paper takes the structure elements and segmented objects as the MRF nodes and propose an indoor scene object segment algorithm base on MAP-MRF. The selection and design of the object nodes is essential for the effective cognition of the objects. This paper takes height as the basic feature for the object. To describe the object more accurately, we propose a shape histogram to represent the 3D shape of the object. Although 3D laser data describe the 3D information, it lacks image and color information. To utilize the image from camera precisely, we fuse the information of 3D laser scanner and monocular vision in pixel level. Under HSI color space, we build color histogram of the object by computing the color distribution of the object points. As the result of scene cognition, the semantic map can be constructed by mobile robot autonomously. Experiment results implemented on real mobile robot platform show the validity of the proposed method.