Study Of Pedestrian Detection Based On LBP And Phase Congruency

Pedestrian detection is the first step for vehicle driver assistance, intelligent video surveillance and human behavior analysis, so it has been a hot topic in the field of computer vision. However, owing to the large variations in pedestrians’ pose and clothing, as well as the varying background and illumination, it’s still a challenging task. Currently the most common method used to detect pedestrian is based on statistical classification method. The core mission of this method is to choose good features and classifiers. Features should have this quality:feature values of samples from the same class should be very similar, and the feature values of sample from different classes should be very different. Classifiers should have good learning and generalization abilities to get the global optimal solution. The statistical classification method extracts the representative features from a great many training samples and then trains a relevant classifier. The method is also adopted in this thesis.The powerful phase congruency feature is introduced in this thesis, Phase congruency feature can tell us the contour information of pedestrian. In order to more effectively describe the characteristics of pedestrians, this thesis takes advantage of image texture information to represent human by Local Binary Pattern (LBP) feature. We combine the phase congruency feature with LBP feature to obtain a new pedestrian detection algorithm and choose support vector machine (SVM) to train the pedestrian classifier. Due to the support vector machine algorithm is a convex optimization problem, so the local optimal solution must be globally optimal solution, it can prevent over-learning.When testing large-scale pictures, we zoom in and out the detection window and slide it on the entire picture. Generally multi-results might be generated on one single object, this thesis integrate all of the test results to get the final result.In order to verify the performance of the new algorithm, a lot of experiments were conducted, experimental results show that the proposed detection algorithm has some improvement in the detection speed and accuracy. The proposed detection algorithm could largely tackle the varied human poses and complex background problem.