To our best knowledge, this paper is the first one which deals with the detection of a person who lies or falls down in the middle of the driving road. To solve the above problem, we propose a novel fallen person image synthesis framework in this paper. However, when a detection network trained on the CityPersons dataset is applied to FPD, the network misses most of the fallen people, as shown in Fig. 1(a), and its detection accuracy degrades to 21.0% (average precision, AP Salton & McGill, 1983) on our RealFPDK1.4K dataset, because the off-the-shelf pedestrian datasets consist of only standing, walking, or running people. Obviously, one might think that we can simply train a detection network using an off-the-shelf pedestrian dataset such as CityPersons (Zhang et al., 2017) which include only standing, walking, or running persons, and apply it to FPD. The difficulty clearly explains why research has rarely been conducted regarding FPD.
Home designer architectural 2017cut and paste how to#
The biggest difficulty in FPD on driving roads is how to capture a sufficient number of training images of people who lie on driving roads because it is quite dangerous. The research regarding FPD on driving roads, which aims at being applied to autonomous driving, has rarely been conducted. Several similar studies have been conducted but they were all restricted to the applications in indoor environments, such as recognizing anomalous actions of the elderly in houses (Lezzar et al., 2020, Maldonado-Bascon et al., 2019, Souza et al., 2020). We refer to this problem as a fallen person detection (FPD) on driving roads. Thus, autonomous vehicles (AV) should definitely have the ability to detect such persons. Experimental results demonstrate that our framework contributes significantly to training an FPD network.Īlthough it rarely happens that a person lies or falls down in the middle of the driving road, it does happen as in Phillips (2022). 210 on RealFPDK1.4K and RealFPDY1.1K, respectively. Our approach achieves AP scores of 0.815 and 0.753, and the scores are higher than those of the baseline by + 0. We verify the effectiveness of our training image synthesis method by applying the detector to the RealFPDK1.4K and RealFPDY1.1K datasets. We released this dataset for the benefit of the autonomous driving society. Our dataset covers a variety of conditions, including occlusion, lack of lighting, and shadows, thereby facilitating qualitative and quantitative evaluations in the real world. The two test sets consist of 14 images of real fallen persons on the road with bounding box annotations, respectively. RealFPDK1.4K and RealFPDY1.1K are test sets which are captured at two different places (K-City and Yonsei University). The two sets will be used to synthesize the driving road images including fallen persons. RealFP218 consists of 218 images of real fallen persons and their pixel-level mask annotations and RealD1.8K consists of 1820 real driving road images. FPD-set consists of four subsets: (1) RealFP218, (2) RealD1.8K, (3) RealFPDK1.4K and (4) RealFPDY1.1K. Furthermore, we develop a new dataset named FPD (Fallen Person detection with Driving scenes)-set to train a detection network. Our proposed framework addresses the lack of training data, which is a serious problem inherent to FPD. Finally, we remove some pixel artifacts from the border between the fallen person and background area in the synthesized image. We then reduce the domain gap between the two images using domain adaptation.
Our framework first embeds a fallen person instance into an image of a driving road, thereby generating a hard-to-acquire image (image of a person who has fallen on a road) from two easy-to-acquire images (driving road image and fallen person image). In this paper, we propose a novel fallen person detection image synthesis framework to address this difficulty. The biggest difficulty in FPD is capturing a sufficient number of training images of people lying on driving roads because of the dangers involved. Fallen person detection (FPD) is a new problem that aims to detect a person who lies or falls down on driving roads.
0 kommentar(er)
