Pedestrian Detection For Advanced Driver Assistance Systems

At the beginning of21th century, traffic accidents have become a majorproblem not only for developed countries but also for developing ones.Advanced driver assistance systems, and pedestrian systems based onComputer Vision, are becoming a strong topic of research aimed at improvingthe safety of pedestrians. On-board pedestrian detection needs processingscenarios from a mobile platform, which implies environments changequickly.However, the challenge is of considerable complexity due to the varyingappearance of pedestrians, the dynamic nature of on-board systems and theunstructured moving environments that urban scenarios represent. Also, therequired performance is demanding both in terms of computational time anddetection rates. There are many ways for pedestrian detection in AdvancedDriver Assistance Systems. However, most approaches based on machinelearning use a large number of features which need much computing time. Toimprove the problem we provide a improved camera pose estimation with3Dspace information analysis method for adaptive sparse image sampling, and aclassifier based on Haar-like wavelets and Real AdaBoost as learningmachine, also a classifier based on HOG and SVM as learning machine.Adaptive road scanning with3D space information analysis method hasfiltered many windows which doesn't contain a pedestrian. And it does twobenefits. First, it decreases computing time in classification process. On theother hand, the detection get less false positives. At last we compare ourproposal with relevant approaches, the results show that our method reducesprocessing time much a lot for the image sampling.