Autonomous vehicle image recognition algorithms analysis and utilization

The thesis is very relevant since vehicle autonomous systems occupying more and more space in people‘s lives. Autonomous systems improves and allows people to entrust bigger part of work for autonomous systems with computer vision. The algorithms are one of the most important things of these systems, it iš important to choose the one that is most effective and provides the best results. The most modern algorithms are based on the features, which are determined by the object. The most important object categories of autonomous vehicles are: other vehicles, road signs, traffic lights, pedestrians. Algorithm for each object group should be chosed individually, taking into account the characteristics of the object. In the thesis it has been discussed the algorithms of a vehicle, traffic signs and traffic lights. For a vehicle detection experiment Haar-Like classifier method was used. The objects of Haar-Like classifier training were divided into 4 groups: • From the front • From the back • From the edge • From the side It has also been analyzed the method dependence on these parameters: • Positive objects • Negative number of pictures • Number of stages For these experiments were used Visual Studio 2015 Community environment together with open source library OpenCV wrapped named EmguCV, as well as the OpenCV open source programs