人群行为分类数据库--Novel Dataset for Fine-grained Abnormal Behavior Understanding in Crowd

Novel Dataset for Fine-grained Abnormal Behavior Understanding in Crowd 数据库：https://github.com/hosseinm/med

本文针对人群行为分类建立了一个数据库，这里有5类：Panic，Fight，Congestion，Obstacle ，Neutral

目前已有的数据库情况：

2 Proposed Dataset

The introduced dataset consists of 31 video sequences in total or as about 44,000 normal and abnormal video clips. The videos were recorded as 30 frames per second using a fixed video camera elevated at a height, overlooking individual walkways and the video resolution is 554 * 235. The crowd density in the scene was changeable, ranging from sparse to very crowded.

3 Proposed Benchmark

这里我们在新数据库上测试了两类算法：

Dense trajectories [25,26](SeeFig. 4)which have shown to be efficient for action recognition are applied as first method on our dataset.

As second benchmark, we use Histogramof OrientedTracklet (HOT) [14, 15,16] descriptor, which is suitable for the task of abnormality detection

Emotion-Based Crowd Representation for Abnormality Detection 这篇文献主要是在这个数据库基础上加入 人群表情信息来 辅助 人群行为的分类。