The main contributions of this research work remain in object recognition by computer vision, by one side, and in robot localisation and mapping by the other. The first contribution area of the research address object recognition in mobile robots. In this area, door handle recognition is of great importance, as it help the robot to identify doors in places where the camera is not able to view the whole door. In this research, a new two step algorithm is presented based on feature extraction that aimed at improving the extracted features to reduce the superfluous keypoints to be compared at the same time that it increased its efficiency by improving accuracy and reducing the computational time. Opposite to segmentation based paradigms, the feature extraction based two-step method can easily be generalized to other types of handles or even more, to other type of objects such as road signals. Experiments have shown very good accuracy when tested in real environments with different kind of door handles.
With respect to the second contribution, a new technique to construct a topological map during the exploration phase a robot would perform on an unseen office-like environment is presented. Firstly a preliminary approach proposed to merge the Markovian localisation in a distributed system, which requires low storage and computational resources and is adequate to be applied in dynamic environments. In the same area, a second contribution to terrain inspection level behaviour based navigation concerned to the development of an automatic mapping method for acquiring the procedural topological map. The new approach is based on a typicality test called INCA to perform the so called loop-closing action. The method was integrated in a behaviour-based control architecture and tested in both, simulated and real robot/environment system. The developed system proved to be useful also for localisation purpose.