The Shenzhen University (SZU) Depth Pedestrian Dataset was created for pedestrian detection research in depth images, and is free of charge for researchers. SZU Depth Pedestrain dataset contains depth images and corresponding intensity images. The depth camera was a Time-Of-Flight camera (TOF camera) designed by Mesa Imaging AG, and the type of the camera is SwissRanger SR4000. Beside depth images, the camera can also capture synchronized intensity images. The resolutions of the depth images and intensity images are all 176 x 144 pixels. The depth images is distance arrays, and the distance is ranging from 0 to 5 meters. The intensity images are gray scale images, the valid range of intensity data is from 0 to 65535.
Using the TOF camera, we captured 4637 pedestrian images and 198 non-pedestrian images in 3 different indoor environments. Pedestrians in these images are all standing or walking. The body orientation to the camera is not limited and can be any direction. The pedestrian positions were manually labeled by rectangles. The 4637 pedestrian (positive) samples are divided into two parts for training and testing respectively. There are 3160 positive training samples, and 1477 positive test samples.
Sample images from the dataset. The first row contains depth images, and the second row contains intensity images.
Related papers:
- Yan-Ran Li, Shiqi Yu*, Shengyin Wu. “Pedestrian Detection in Depth Images Using Framelet Regularization“, the 2nd International Conference on Computer Science and Automation Engineering, 2012.
- Shengyin Wu, Shiqi Yu*, and Wensheng Chen. “An Attempt to Pedestrian Detection in Depth Images“, the 3rd Chinese Conference on Intelligent Visual Surveillance, 2011.