CrowdFlow (TUB CrowdFlow)

数据集包含10个序列，显示5个场景。每个场景渲染两次: 使用静态视角和动态摄像机来模拟基于无人机/无人机的监视。我们以25 fps的高清分辨率 (1280x720) 渲染，这对于当前的商业闭路电视监控系统来说是典型的。帧的总数3200。
对于每个序列，我们提供以下地面真实数据: 光流场，人轨迹 (最多1451)，密集像素轨迹。

This dataset comprises 10 sequences representing 5 distinct scenarios. Each scenario is rendered twice: once with a static camera perspective and once with a dynamic camera to simulate unmanned aerial vehicle (UAV)/drone-based surveillance. All sequences are rendered at high-definition (HD) resolution (1280×720) at 25 frames per second (fps), a frame rate typical of contemporary commercial closed-circuit television (CCTV) surveillance systems. The total number of frames across all sequences is 3200.
For each individual sequence, the following ground truth data is provided: optical flow fields, human trajectories (with a maximum count of 1451), and dense pixel-wise trajectories.

搜集汇总

背景与挑战

背景概述

CrowdFlow数据集包含10个序列，覆盖5个不同场景，每个场景均以静态和动态摄像机视角渲染，模拟无人机监控环境。它以25fps的高清分辨率（1280x720）提供3200帧，并附带光流场、人轨迹和密集像素轨迹等地面真实数据，适用于视觉人群分析任务。