In the thesis we investigated the problem of environmental sensing with a stereo camera, which is commonly used in the field of mobile systems for three-dimensional sensing of the environment in the form of a point cloud. The problem of reconstruction is usually solved by stereo vision based on epipolar geometry. We use three-dimensional information to build a local robot map based on which we implemented collision avoidance algorithms.
With the D435i depth camera, we obtain a depth image, which was used for reconstruction of a three-dimensional representation of the environment through triangulation. With the segmentation of the environment we built a local map of the robot, where we improved the representation of the obtained map of the robot by means of a morphological dilation operation and a hole filter. On the map, segmentation achieved the division of space into floors, obstacles and unoccupied areas, i.e. areas where we were unable to retrieve information from the current scene. Based on this information, we implemented a system for preventing collisions when driving straight and when cornering.
Collision prevention is based on a proportional reduction of speed with respect to the distance to the closest obstacle or minimum time to impact. In situations that may result in a collision with an obstacle, the control system shall intervene and, if necessary, bring the mobile system to a standstill.
In the master's thesis we tested obstacle detection with a stereo camera and the operation of control algorithms on a Pioneer 3-AT robot, which is commonly used for research purposes. Obstacle detection and collision avoidance systems meet the requirements for real-time operation. The algorithms are implemented in the Matlab software tool, which communicate with robot via a wireless connection. A meta-operating system ROS is running on the robot.
In this thesis, we introduced the calibration of an obstacle detection system that allows the robot to determine automatically the ground plane and pose of the stereo camera in the plane. This kind of calibration is also important in order to simplify the work with the mobile system (e.g. in the industry), since the possibility of automatic calibration can eliminate the need for expert knowledge and reduce the cost of system integration.