Image Based Vehicle-Trailer Angle Estimation

Abstract

A trailer assist system is an Advanced Driver Assistance System. It enables a more effective maneuvering of a trailer towed by a vehicle. It is particularly useful in constraint scenarios such as parking, traffic jam, turns, etc. Traditionally, trailer assist systems are based on image processing methods or require specific sensors to measure the hitch angle. In this paper, we propose to estimate the hitch angle by making use of the existing rear camera of the vehicle and by using an image processing approach, thus avoiding the supplementary angle measurement sensor. To this extent, a dedicated database is collected and annotated. Two different methods are developed to predict this corresponding angle, the first method is based on a convolutional neural network and the second one is a deterministic image processing method. The two approaches are finally compared in terms of accuracy and execution time.