Intelligent Park assists system: Pedestrian detection based on fisheye camera

Pedestrian detection is a key problem in computer vision, with several applications including surveillance and automotive safety. Advanced vehicles equipped with camera vision based advanced driver assistance systems (ADAS) increase rapidly. A single fish-eye camera is used in

order to get much larger FOV (field of view) for object detection. Safety and ease-of-use of vehicles are the two central themes in this line of work. This paper focuses on the safety aspect and presents a new prototype system for the detection of the most vulnerable traffic participants: pedestrian using a fisheye mounted camera. This paper proposes a robust rear and front view camera based moving object detection algorithm for backup aid and parking assist applications. A real time vision-based system for driver assistance on-board vehicles based on distorted image is developed. The dataset contains richly annotated video, recorded directly from a vehicle, with challenging images of fisheye camera. The main components of a pedestrian detection system and the underlying models are covered. A diverse set of state-of-the-art systems: Haar waveletbased AdaBoost cascade, HOG, SVM and other combined detection are considered. Experimental results under various conditions show that most of moving pedestrians can be detected and classified correctly by our algorithm.