Research On Recognition Algorithm Of Crosswalk And Diamond

With the city heavier traffic, traffic safety has increasingly become a serious social problem. Accidents occur at the intersection with the relatively high rate, and bring the pedestrian and vehicle safety into a great threat, so it is helpful to predict the intersection in reducing traffic accidents. Therefore, the study in crosswalk and diamond identification algorithm has been identified with the notice and has been an important direction of intelligent transportation.So far, the research of crosswalk recognition algorithm is mainly used in blind navigation systems, and rarely applied to the study of intelligent transportation systems, meantime, identifying the markings in difficult lighting scenarios such as the markings blocked, damaged and impacted with the different lightings, which is still the bottleneck problem. The research of crosswalk and diamond recognition in this article mainly talks about the crosswalk recognition technology and mainly used in intelligent vehicle navigation system. What's more, marking recognition algorithm for the difficult scenarios is being studied and also improves the recognition results in difficult situations.This thesis presents a crosswalk and diamond recognition method by using a monocular camera on a moving platform. The recognition algorithm is divided into two steps:marking candidate region extraction and verification. Extraction stage of the crosswalk and diamond candidate region, an algorithm to extract candidate regions based marking highlight slices filtering is proposed. In this process, firstly, an Inverse Perspective Mapping (IPM) transformation is introduced to suppress the perspective effect in the image; Secondly, the road marking candidate region is extracted based on high brightness slice filtering with Start-End Point Finding algorithm adopted on the platform line by line, to solve the problem with the traditional single threshold to extract the candidate region of the markings casted with the shadows and impacted. Furthermore, we propose the sliding window searching method based on the prior knowledge of the markings to generate marking candidate region with the marking-slices. To verify the candidate areas precisely, a coarse-to-fine recognition algorithm is proposed. The design combining the Haar characteristics with the Adaboost cascade classifier achieves a coarse filter of marking candidate region, and filters out marking and non-marking candidate region. Then, an improved ELM with an adaptive selection of parameters adopted, combines the HOG features selected based on the BW algorithm to verify the marking types and background, and to give the recognition results.In this thesis, the algorithm is achieved and tested by using a monocular camera on the front of the vehicle. The results show that the algorithm can accurately and quickly identify the crosswalk and diamond, and also can effectively deal with the blocked marking, damaged markings and in other scenarios, which can improve the recognition results.