Obstacle detection and planning for autonomous vehicles based on computer vision techniques

Résumé

In this thesis, we propose different methods which are used for the detection of obstacles in the environment of our prototype, Verdino. In the first part of this document, we describe a set of methods thought for this purpose, each of these making use of different techniques. This allows comparing the advantages and disadvantages of each of them, from the point of view of the proposed application. The requirements for the obstacle detection method are: getting a reasonable obstacle detection rate using as input the images provided by the cameras on board; ability to work with non static cameras; ability to localize the obstacles in the map; real time performance; independence on the environment conditions and tracking capabilities. Once obstacles are detected, we need to avoid them. In the second part of the thesis, we describe the method used by the prototype to travel around the map. This method comprises two tasks: the first one computes a trajectory that starts in the position of the vehicle and ends in the target to which we want to move. If we had a Road Network Definition File (RNDF), we would not need to compute this trajectory, but since Verdino travels in an unstructured environment, this step is needed. In the second step, Verdino does the calculations needed to follow the computed trajectory. In this stage, we want the vehicle to be able to follow the track but, at the same time, we also want to have enough flexibility in the commands generated to ensure that it is able to abandon the track if necessary in case we want to avoid an obstacle. At the end of the document, some results corresponding to the studied methods are shown, as well as a discussion of the advantages and disadvantages of each of them. Some conclusions are described and future work is proposed.