A Novel Detection Framework for Detecting Abnormal Human Behavior

Table 1

Introduction of data sets.

Data set	Details
UCSD [26]	The database contains two video sets, Ped1 and Ped2. The Ped1 video set contains a total of 34 training video sequences and 36 test video sequences, each with a resolution of 238 ∗ 158. Each video sequence contains 200 frames of images. The Ped2 data set contains a total of 16 training videos and 12 test videos, each with a resolution of 320 ∗ 240.
UMN [27]	The database contains 3 crowded scenes; the total numbers of frames in the three scenes are 1450, 4415, and 2145, and the resolution of each is 320 ∗ 240.
Subway [28]	The database contains the subway-entrance and subway-exit subsets. The video lasts 1 hour and includes abnormal behaviors such as retrograde and fare evasion at the ticket gate.
Avenue [29]	The data set has 16 training videos and 21 test videos, each with a resolution of 640 ∗ 360. Abnormal behaviors include throwing paper everywhere and running.