| 초록 |
This paper proposed a method to perform parking slot detection and segmentation simultaneously through a deep neural network that uses around view monitor(AVM) images. The proposed method used an end-to-end, trainable, one-stage convolutional neural network(CNN) to detect locations, orientations, types, and occupancies of parking slots while performing segmentation on parking slot markings. Moreover, this paper suggested an alternating training method that could effectively train a single network to perform two tasks(detection and segmentation) in situations with a label imbalance problem between the detection and segmentation datasets. In experiments, the proposed method was quantitatively evaluated by using two public AVM image datasets. Results confirmed that the proposed model, which performed both detection and segmentation simultaneously, exhibited superior or similar performance compared to models that performed either detection or segmentation only. |
