A trend of the SLAM (Simultaneous Localization and Mapping) researches is to use low cost video cameras instead of using expensive specialized sensors to capture 3D information from the environment. This approach is usually referred to as Visual Slam or Visual Odometry. The choice of this type of this solution relies on the fact that there is a wide availability of video cameras and they can even be used in situations where other sensors fail. However, the extensive use of Visual Odometry still faces various performance and robustness problems. This project aims to research and develop Visual Odometry and object tracking in image sequences, within the autonomous robot navigation context. The study of feature point extraction algorithms recently presented in the literature will be emphasized, seeking to benchmark performance and robustness issues in the context of a specific application. Thus, the work involves the development of an application of Visual Odometry for movement control of a mobile robot. As a result of the conducted research it is expected the understanding about needed concepts for spatial recovery models development that are appropriated to the autonomous robot navigation.
News published in Agência FAPESP Newsletter about the scholarship: