Simultaneous Localisation And Mapping (SLAM) methods have been developed with the aim of automating robot navigation. Place recognition is considered a key element, complementary to sequential motion estimation done in visual SLAM, to enable global accurate maps, relocalization and even collaboration between different robots. Typically, 2 main tasks must be accomplished to solve the place recognition problem: a) query a database of images in order to find possible similar locations and then b) determine which, if any, of these images represent the same place as the query. Usually, a Bag of Words approach together with an inverted file index is used in the first task followed by a geometric consistency check in the latter one. Considering place recognition onboard of UAVs is especially challenging given that a UAV can revisit a place from very different viewpoints. Moreover, different scenes exhibit different properties and challenges for place recognition -- e.g. environments with little or much texture, planar/non-planar scenes and scenes being revisiting from a very different viewpoint. In order to improve a traditional appearance-only place recognition a suitable geometric verification need to be applied. Suitable geometric tests enable more robustness to different kind of scenes and increase viewpoint tolerance. Given the variety of scenes that can be encountered by a UAV, the goal of this project is to develop a method to tackle the problem of identifying previously visited places for different kind of environments. The objective here is to develop an algorithm to distinguish among different types of scene and apply an appropriate geometric check for each kind of environment in order to improve a place recognition system.
The student will work with a real setup and equipment offered by V4RL. This work is part of a large European project (AEROWORKS) and a successful method will directly be used within in the framework developed for this project.
Simultaneous Localisation And Mapping (SLAM) methods have been developed with the aim of automating robot navigation. Place recognition is considered a key element, complementary to sequential motion estimation done in visual SLAM, to enable global accurate maps, relocalization and even collaboration between different robots. Typically, 2 main tasks must be accomplished to solve the place recognition problem: a) query a database of images in order to find possible similar locations and then b) determine which, if any, of these images represent the same place as the query. Usually, a Bag of Words approach together with an inverted file index is used in the first task followed by a geometric consistency check in the latter one. Considering place recognition onboard of UAVs is especially challenging given that a UAV can revisit a place from very different viewpoints. Moreover, different scenes exhibit different properties and challenges for place recognition -- e.g. environments with little or much texture, planar/non-planar scenes and scenes being revisiting from a very different viewpoint. In order to improve a traditional appearance-only place recognition a suitable geometric verification need to be applied. Suitable geometric tests enable more robustness to different kind of scenes and increase viewpoint tolerance. Given the variety of scenes that can be encountered by a UAV, the goal of this project is to develop a method to tackle the problem of identifying previously visited places for different kind of environments. The objective here is to develop an algorithm to distinguish among different types of scene and apply an appropriate geometric check for each kind of environment in order to improve a place recognition system. The student will work with a real setup and equipment offered by V4RL. This work is part of a large European project (AEROWORKS) and a successful method will directly be used within in the framework developed for this project.
- WP1: Research into existing works tackling the problem of place recognition and SLAM.
- WP2: Development of a new algorithm to identify different kind of scenes and their respective geometric consistency check.
- WP3: Experimentation and evaluation of this method in terms of runtime and accuracy of estimation.
- WP4: Integration of the proposed method into a state-of-the-art odometry system.
- WP5: Experimentation and final evaluation of the method with respect to the state-of-the-art and report writing.
- WP1: Research into existing works tackling the problem of place recognition and SLAM. - WP2: Development of a new algorithm to identify different kind of scenes and their respective geometric consistency check. - WP3: Experimentation and evaluation of this method in terms of runtime and accuracy of estimation. - WP4: Integration of the proposed method into a state-of-the-art odometry system. - WP5: Experimentation and final evaluation of the method with respect to the state-of-the-art and report writing.
- Background knowledge in computer vision.
- C++ programming experience.
- Experience with Linux, ROS are advantageous.
- Strong self-motivation and critical mind.
- Background knowledge in computer vision. - C++ programming experience. - Experience with Linux, ROS are advantageous. - Strong self-motivation and critical mind.