Improved HOG Pedestrian Detection Means Using V-disparity Map

In order for a driverless car to drive safe, it is important to accurately estimate information about any obstacles near the car and especially the existence and position of pedestrians. A histogram of oriented gradient (HOG) support vector machine (SVM) is widely used in pedestrian detection algorithms but can cause a false positive problem. This paper will suggest an improved pedestrian detecting means that is using Euclidean distance of coordinate values of the center in extracted foreground objectives after removing ground and background and in the extracted HOG based SVM from an input sequence through V-disparity map to achieve safe pedestrian detecting in dynamic conditions. Its superior performance compared to the existing HOG-based SVM was proven through the results of an image sequence.