With the rapid development of self-driving cars, automatic parking technologies have been widely concerned. However, extreme conditions such as different illuminations and incomplete features bring huge challenges for the parking slot detection, which is a key to automatic parking systems. To accurately and quickly detect parking slot features in such unfavorable conditions, an efficient park slot feature detection method based on the Convolutional Neural Network (CNN) is proposed in this paper. We collect simulated parking slot images under various extreme conditions which are taken as the input of the network as dataset. For each image in dataset, the parking slot feature points are carefully labeled. The YOLO v3 is applied as the basic neural network framework and the transfer learning is employed to train the network so as to accurately detect the parking slot features. Experimental results show that the parking slot feature recognition accuracy of the proposed method exceeds 96%, and the detection speed reaches 26 Frames per Second (FPS).
Keywords Parking slot feature detection, Transfer learning, Convolutional neural network