The APT dataset is composed of four parts: Video Trajectoris Object types Coherence qualities ==================================================================================================== Video: Recorded at 29 fps with a frame size of 720*480 for a duration of 30 minutes using a Sony Handycam HDR-XR520V. ==================================================================================================== Trajectories: For ease of processing, the trajectories computed in time windows of 2 minutes. Thereby there are 15 files in this folder, each of which relate a 2 minute time window. The location information is derived from the range readings of 4 Hokuyo UTM-30LX sensors positioned at the coners of 8.5mx7m rectangular area. The view angle of the sensors are [-90, +90] and the sampling rate is 720/scan. For the tracking and localization algorithm see: D. F. Glas, T. Miyashita, H. Ishiguro, and N. Hagita, “Laser-based tracking of human position and orientation using parametric shape modeling,” Advanced Robotics, vol. 23, no. 4, pp. 405–428, 2009. The first column defines unix time stamp, where the second column keeps the tracking id. The third and fourth columns give the x- and y- locations of the concerning object, respectively. The location values are given in milimeters. ==================================================================================================== Coherence qualities: For ease of processing, the coherence qualities are annotated in time windows of 2 minutes. Thereby there are 15 files in this folder, each of which relate a 2 minute time window. The first column defines the nature of coherence relation, where the second and third columns contain the tracking ids of the concerning coherent objects. The coherence relation is denoted with numbers 0-5, according to the following mapping: 1: Human-human pair 3: Shopping cart-human pair 4: Small baby cart-human pair 5: Big baby cart-human pair In case of cart-human pairs, the id of the cart is given in the second column and the id os the human is given in the third column. For the definition of coherence quality see: Yucel, Z., Zanlungo, F., Ikeda, T., Miyashita, T., Hagita, N., Modeling Indicators of Coherent Motion, EEE International Conference on Robotics and Automation, ICRA 2012. ==================================================================================================== Object types: For ease of processing, the object types are annotated in time windows of 2 minutes. Thereby there are 15 files in this folder, each of which relate a 2 minute time window. The first column is the tracking id and the second column is correspoding object type. The types are denoted with numbers 0-5 according to the following mapping: 0: Undefined 1: Human 2: Wheelchair 3: Shopping cart 4: Small baby cart 5: Big baby cart ====================================================================================================