Localization of Pedestrian with Respect to Car Speed

Pedestrian location with respect to the road and its distance from vehicle are important information for driver assistance system. This work proposes a method to classify location of pedestrians using estimated road region and vehicle movement region. Single camera is used to detect pedestrians, thus classifying their locations to assist the driver in avoiding accidents. The system consists of three stages. First, moving pedestrians are detected using optical flows method. They are extracted from the relative motion by segmenting the region representing the same optical flows after compensating the ego-motion of the camera. Thus, histogram of oriented gradients (HOG) features descriptor and linear support vector machine (SVM) are used to recognize the pedestrian. Second, road lane is detected using combination of color based and Hough based detection method. It is used to define road boundary region by connecting left and right lanes on a vanishing point. Third, a heuristic method according to a vehicle movement region is defined by the typical stopping distance. It is depended on the speed of the car that will be impact to the thinking distance and braking distance. Combination of estimated road region and vehicle movement region are used to classify the location of the detected pedestrians. They are classified into three zones; car movement zone, road zone surrounding car path and the pedestrian track zone. The proposed method is evaluated using ETH and Caltech datasets, and the performance results shown the best pedestrian detection rate is 99.50 % at 0.09 false positive per image. The location classification is applied in our real world driving data, the evaluation result shows correct classification rate is 98.10 %.