Berlin-APC: A Privacy-Friendly Dataset for Automated Passenger Counting in Public Transport
Seidel, Robert; Zarafeta, Dimitra; Siebert, Michael; Dastgheib Shirazi, Ramtin; Seo, Sambu; Goerttler, Thomas; Obermayer, Klaus
This document provides a short technical introduction to the Berlin-APC dataset. The dataset consists of two files, a HDF5 file which contains the image sequences, and a CSV file which contains the labels. The CSV file has three columns: (1) the image sequence name; (2) the number of boarding passengers in that image sequence; (3) the number of alighting passengers in that image sequence. The image sequence names also serve as keys in the HDF5 file. The HDF5 file’s datasets (indexed by the aforementioned sequence names) are float16 arrays of the shape (number of frames, 20, 25), the pixel values range between 0–1, and the frame rate is 10 frames per second.
