![]() ![]() This device is able to provide GPS localization with an error below 3 meters. The second hardware element is the GlobalSat GPS device. This range is limited by the baseline between the two cameras featured in the device. This camera is able to sense depth values ranging between 0.7 and 20 meters. This setting limits the framerate to 15 fps. In order to provide better image quality, we set the resolution to the maximum, namely 2208 × 1242 resolution. It supports a number of different resolutions and framerates. The ZED is a 2 K stereo device with dual 4 MP RGB sensors, a 110° field of view and f/2.0 aperture. First, the images and the depth maps were captured by a ZED camera. Two main hardware devices were used for the creation of the dataset. ![]() These features lend the dataset high variability, which will challenge the generalization capabilities of the algorithms (see Technical Validation section). It also involved up to 4 different persons filming the dataset at different moments of the day. The dataset provides a GPS geolocalization tag for each second of the sequences and reflects different climatological conditions. It provides about 160902 frames, thus yielding sufficient data to be able to train a deep learning network. Another important feature is the scale of the dataset. The images were extracted from video files with 15 fps at HD2K resolution with a size of 2280 × 1282 pixels. The frames show different paths from the perspective of a pedestrian, including sidewalks, trails, roads, etc. It was created at the University of Alicante and consists of an RGB-D stereo dataset, which provides 33 different scenes, each with between 2 k and 10 k frames. The dataset presented in this paper is UASOL 8: A Large-scale High-resolution Outdoor Stereo Dataset. It only contains 534 frames, so the scale of this dataset is the smallest of all those reviewed. The different scenes provide human interaction and also different types of paths and roads a pedestrian could use. In addition, the amount of data is not enough to correctly train a more complex deep learning algorithm.įinally, the Make3D dataset 3 is outdoor and taken from the perspective of a pedestrian. The Middlebury dataset provides different lighting conditions for each scene, but as mentioned, theseare indoor static scenes. All of the scenes provided are indoor, mainly focused on objects. The Middlebury Dataset 7 provides 33 scenes, each filmed from two different exposures. They provide only 6 scenes with complete rooms, which are not ordinary scenes and therefore could not provide good generalizability to the trained models. It only provides static scenes with no interaction, which leads to the scenes provided are being mostly objects. The ground-truth data was captured using an industrial laser scanner which adds precision to the data. Tanks and Temples 6 includes 147791 RGB-D frames in 14 different scenes. ![]() Of the 25 scenes, only 9 are outdoor which significantly decreases the number of images. The ground-truth was taken with a highly accurate 3D laser scanner. The ETH3D dataset 4 includes 534 RGB-D frames divided into 25 scenes. The only problem of this dataset is that it was captured from the perspective of a car, so the main view is from the road. This dataset is outdoor, so it fulfills one of our main requisites. The KITTI dataset 5 provides the RGB (stereo pair) and depth maps of 400 different layouts having a total of 1.6 k frames of roads from the city of Karlsruhe (Germany). The second problem is that the dataset is centered on the vision of a car driving in the street, while, in our case, we need the point of view of the pedestrians. The first is that it is a synthetic dataset, meaning that the frames are not photorealistic, with the subsequent problem of testing the system in real conditions. There are two main problems with this dataset. SYNTHIA 2 or The SYNTHetic collection of Imagery and Annotations consists of a collection of photo-realistic frames rendered from a virtual city. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |