The project “The generalized, multi-robot framework for an augmented visual simultaneous localization and mapping system” was finished on the 7th of April. The project was financed by the Polish National Science Centre within the PRELUDIUM program according to the decision DEC-2011/01/N/ST7/05940. The total budget of the project equalled 125760 PLN.
The main contribution of the project is the development of the generalized, multi-robot visual SLAM framework. The proposed framework facilitates fast design of the visual SLAM systems composed of arbitrary number of robots, cameras and markers according to the top-down design paradigm. Moreover, the framework enables the evaluation of different variations of the visual SLAM systems on a common ground and can be easily extended to incorporate new models, techniques and algorithms.
The second contribution is the design of new robot’s movement model enabling the measurements of the robot’s orientation change and thus improving the accuracy of the trajectory reconstruction. Moreover, the influence of the movement parametrization on the system’s performance was evaluated.
Thirdly, two simplified representations of the point features and local maps were introduced. The reduced models not only significantly reduce the dimensionality of the SLAM problem but also improve the accuracy of the tracking. Additionally, the introduction of the local maps enabled vision-based measurements of the relative orientation between the maps and the moving cameras. Moreover, the introduction of the local maps allows for predicting the appearance of the point feature’s neighborhood in the image from the current camera pose and warping the template, thus improving the feature matching.
Fourthly, the environment model was extended with static cameras and artificial markers installed both statically and on the moving robots. The additional measurement sources improved the robot’s trajectory tracking. The mechanisms for initialization of the multi-robot system and for mutual robot observations were proposed. Moreover, an efficient method for detection and identification of the marker as well as the calculation of its pose w.r.t. observing camera’s coordinates was designed and evaluated.
Fifthly, the thorough evaluation of the popular image feature point detectors and descriptors in the contexts of both the essential matrix estimation and visual SLAM was performed.
Finally, the robot movement capture system was designed and constructed. The methods for calibration of the spatial relations between the system’s cameras as well as the cameras and markers placed on the robots were designed. The system was used to register unique experimental data facilitating the work on single- and multi-robot visual navigation. All the proposed modifications of the visual SLAM system were evaluated using the newly obtained data. The registered video sequences along with the reference trajectories are publicly available to facilitate further research on visual robot navigation .
The results of the project were presented on the ACIVS2013 and ICMMI2013 conferences and in the Journal of Automation, Mobile Robotics & Intelligent Systems. One paper was accepted for publication in the International Journal of Advanced Robotic Systems and three more were submitted to the ICCVG2014 conference. The robot motion capture system and the two mobile robots acquired during the project will be used in further research on the mobile robots navigation.