CrowdFlow
收藏资源简介:
CrowdFlow数据集由柏林工业大学通信系统组创建,专注于视觉人群分析中的光流应用。该数据集包含超过3200帧的视频序列,模拟了五种典型的人群分析场景,使用最新的视频引擎生成。数据集旨在解决现有光流数据集在人群行为分析方面的不足,特别是在精确运动估计和长时间范围内的运动一致性方面。CrowdFlow数据集适用于研究人群动态、分割和行为分析,为视觉监控领域的光流算法提供了新的挑战和评估标准。
The CrowdFlow dataset was developed by the Communication Systems Group at Technische Universität Berlin, targeting optical flow applications in visual crowd analysis. Comprising video sequences with over 3,200 frames, the dataset simulates five typical crowd analysis scenarios and is generated using state-of-the-art video engines. It is designed to address the limitations of existing optical flow datasets for crowd behavior analysis, especially in terms of accurate motion estimation and motion consistency over extended time periods. The CrowdFlow dataset is suitable for research on crowd dynamics, segmentation and behavior analysis, and provides novel challenges and evaluation benchmarks for optical flow algorithms in the field of visual surveillance.

- 1Optical Flow Dataset and Benchmark for Visual Crowd Analysis柏林工业大学通信系统组 · 2018年



