A parking assistant system

Abstract

Parking a car can be very cumbersome and difficult in many circumstances, especially when the rear view of the car is blocked. This thesis designs a prototype for an efficient parking assistant system that aims to facilitate the parking process by providing the drivers with distance information of obstacles in the scene. A video camera mounted on the rear window of a car is used to supply a sequence of input images to the system which then generates a 3-D depth map of the environment. The algorithm proposed in this thesis to reconstruct 3-D distance information from an image sequence involves several steps. First, feature points are extracted in each image using a Curvature Scale Space (CSS) based corner detector. Corresponding feature points are then matched between consecutive frames and tracked through the entire sequence. Using the motion parameters of the vehicle obtained from measuring the angular moment of the steering wheel and the distance information provided by the wheel encoders attached to the rear wheels of the vehicle, the distance of objects in the scene can be reasonably estimated. The algorithm is tested on several sets of real image sequences captured inside a parking lot. Some experimental results are presented and analyzed in detail.