Skip to Main content Skip to Navigation

Navigation visuelle d'un robot mobile dans un environnement d'extérieur semi-structuré

Abstract : This thesis deals with the automatic processing of color images, and its application to robotics in outdoor semi-structured environments. We propose a visual-based navigation method for mobile robots by using an on-board color camera. The objective is the robotization of agricultural machines, in order to navigate automatically on a network of roads (to go from a farm to a given field). Firstly, we present an analysis of the main research works about visual-based navigation literature. A preprocessing chain for color rendering on mono-sensor digital images equipped with a Bayer filter, is presented; it is based on the analysis of the demosaicking techniques, the chromatic calibration of images (white point balance) and the correction gamma. Our monocular scene interpretation method makes possible to extract the navigable regions and a basic 2D scene modeling. We propose functions for the segmentation of the color images, then for the characterization of the extracted regions by texture and color attributes, and at last, for their classification in order to recognize the road and other entities of the current scene (grass, trees, clouds, hedges, fields,&). Thus, we use two supervised classification methods: Support Vector Machines (SVM) and k-nearest neighbors (k-NN). A redundancy reduction by using independent components analysis (ICA) was performed in order to improve the overall recognition rate. In a road network, the robot needs to recognize the roads intersections in order to navigate and to build a topological model from its trajectory. An approach for the road classification is proposed to recognize: straight ahead, turn-left, turn-right, road intersections and road bifurcations. An approach based on the road shape representation and categorization (shape context) is used for this purpose. A validation was carried out on an image dataset of roads or country lanes. By exploiting this method to detect and classify the nodes of a road network, a topolo gical model based on a graph is built; the method is validated on a sequence of synthetic images. Finally, Robot displacement is controlled and guided by the information provided by the vision system through elementary displacement primitives (Road-Follow, Follow-Object, Follow-Border,&). Robot DALA is placed in the middle of the road by computing a trajectory obtained from the navigable region contours. As retrieving semantic information from vision is computationally demanding (low frequency 0.5 to 1 Hz), a Snakes tracking process was implemented to speed up the transfer of instructions (5 to 10 Hz) to the locomotion module. Both tasks must be synchronized, so the tracking can be re-initialized if a failure is detected. Locomotion tasks are planned and carried out while waiting for new data from the vision module; the instructions which are not yet carried out, are merged and filtered with the new ones, which provides stability to the system.
Document type :
Complete list of metadata

Cited literature [177 references]  Display  Hide  Download
Contributor : Emilie Marchand <>
Submitted on : Tuesday, November 8, 2005 - 2:20:42 PM
Last modification on : Friday, January 10, 2020 - 9:08:09 PM
Long-term archiving on: : Friday, April 2, 2010 - 11:10:25 PM


  • HAL Id : tel-00010912, version 1


Juan Gabriel Avina Cervantes. Navigation visuelle d'un robot mobile dans un environnement d'extérieur semi-structuré. Automatique / Robotique. Institut National Polytechnique de Toulouse - INPT, 2005. Français. ⟨tel-00010912⟩



Record views


Files downloads