Pedestrian detection for driver assistance systems

Pedestrian detection is an important key problem in Advanced Driver Assistance Systems (ADAS). Un-signalized pedestrian crossing zone are dangerous places, where pedestrians enter the lane suddenly. This is the main factor for most of the accidents. For that, this paper illustrates a machine learning approach for detecting the pedestrian zone and also to detect the pedestrians crossing in that zone. This is implemented by two different stages. In the first stage, the system checks for the presence of the pedestrian zone by combining the advantages of extended Center Symmetric - Local Binary Pattern (XCS-LBP) method and Adaptive Background Mixture Model for Foreground detection. Then it employs the Histograms of Oriented Gradient (HOG) for the most accurate set of features and Linear Support Vector Machine (LinSVM) to classify whether the pedestrians present or not. The reason why the Linear SVM classifier is selected is because SVM provides the high generalization capacity and classifies more effectively. In the second stage, it analyzes the pedestrian crossing event for detecting the pedestrians whom crossing the zone suddenly. This second stage is performed, only if there is a presence of pedestrian is detected in the input video frames. So in this system, it processes only the video frames which contain the pedestrians. Thus, this approach processes the input video frames more rapidly and attains higher detection rates.