In 2011, the National Natural Science Foundation of China released application instructions of the major research plan `Cognitive Computing of Visual and Auditory Information`. In unmanned autonomous vehicles (UAV) research field, a basic research cultivation project named `multi car-like robot interaction and intelligent transportation simulation` was involved in this plan. The project required that `In three-dimensional miniature urban traffic environments, more than 80 car-like robots with different sizes and heterogeneous vehicle groups form traffic flows on traffic intersections and multi-lanes to simulate urban traffic conditions. Therefore, researches on interactive behavior, cognitive model of multi intelligent vehicles, and corresponding control methods can be conducted. Such researches will provide basic theory and technical supports for researches and development of interactions of unmanned intelligent vehicles and urban intelligent transportation.` The methods of traffic sign detection and recognition are basis and one of the key issues of the interactive behavior cognitive model of multi-intelligent vehicles. Their research findings can not only be taken as references for the interactive researches of UAV, but also promote the development of UAV technologies. Compared to UAV, intelligent car-like robots have the advantages of small size, low cost, and short development cycle. However, their load capacity is limited because they are made of plastic materials. Since they can not carry heavier or larger devices, the performance of their loaded image sensor and processors is low. To take the challenge of undesired conditions that the resolutions and quality of images are low and the processing capacity is limited, this thesis is focused on studying methods of traffic sign fast detection and recognition methods. The main contributions are as follows: Usually, it is hard to select an appropriate threshold for color segmentation. In order to address this problem, this thesis proposes a method based on color template mask generated from sample data. Experimental results show that this method has improved accuracy rate of color segmentation. For the problem of occlusion, this thesis designs a shape matching system with two levels based on moment invariants and circle detection. Due to these two level treatments which aim to fast detection of non-occlusion and part occlusion traffic signs respectively, the overall det...
修改评论