Obstacle Detection For Parking Systems With A Rear Camera

Advanced parking assistance systems always play a greater role in the automotive industry and the in the desire to increase the overall safety for passengers and pedestrians.Modern vehicles,amongst other assistance systems,provide a rear-view camera.The generated graphical information of these cameras and the use of machine learning algorithm allow the detection of any desired object,such as parking obstacles.This research work develops four different detectors,which are based on four different training sets.Different types of traffic cones shall be detected with those detectors.The four training sets represent an original training set with three additional levels of augmentation.Afterwards,the influence of those four different detectors on two different test subsets is investigated.As the first test subset meets the same environment conditions as the training set and as the second test subset varies greatly in the environment conditions,a theory is set up,which states,that the detectors score better results on the first test subset.The results show,that the fourth detector scores a maximum average detection rate of 92.39%on the first test subset and that the first detector scores a maximum average detection rate of 75.53%on the second test subset.The results further show,that different augmentation levels of the training set increase the detection rate on the first test subset,but generally decrease the detection rate on the second test subset.As a conclusion,the previously set up theory was proven to be right.