Anomaly Event Detection Based on Two-Stream Network and Multi-instance Learning

Xianbin Yang; Jianwu Dang; Song Wang; Yangping Wang

doi:10.3788/LOP202158.2015006

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 20 >Page 2015006 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 20, 2015006 (2021)

Anomaly Event Detection Based on Two-Stream Network and Multi-instance Learning

Xianbin Yang¹, Jianwu Dang^1、2、*, Song Wang^1、2, and Yangping Wang^2、3

Author Affiliations

¹School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China

²Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphic & Image Processing, Lanzhou, Gansu 730070, China;

³National Experimental Teaching Demonstration Center of Computer Science and technology, Lanzhou Jiaotong University, Lanzhou, Gansu 730070, China

show less

DOI: 10.3788/LOP202158.2015006 Cite this Article Set citation alerts

Xianbin Yang, Jianwu Dang, Song Wang, Yangping Wang. Anomaly Event Detection Based on Two-Stream Network and Multi-instance Learning[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2015006 Copy Citation Text

show less

Fig. 1. Flow chart of the proposed anomaly detection method

Download full size

Fig. 2. ROCs of different anomaly detection methods

Download full size

Fig. 3. Comparison charts of qualitative results for video frame tests of various anomaly events. (a) Explosion; (b) animal abuse; (c) traffic accident; (d) normal video frame fragment

Download full size

Fig. 4. Score variations of anomaly video clips under different iteration times. (a) 2000 times; (b) 5000 times; (c) 8000 times; (d) 12000 times

Download full size

Experimental environment	Detailed information
	CPU: Intel(R) Core(TM) i7-8700 at 3.20 GHz
Computer	RAM: 16 GB
	MATLAB 2017
	Python 2.7
	GPU: NVIDIA GeForce GTX 1060 3GB
Server	TensorFlow 1.6
	Display memory: 12 GB
	Ubuntu 14.04

Table 1. Configuration of experimental environment

Model	Network	AUC	Detection speed / fps
	Time flow: RGB	71.43
I3D	Spatial flow: TV-L1	73.61
	Two-stream fusion model: TV-L1+RGB	76.83	25
	Time flow: RGB	71.43
M-I3D	Spatial flow: MotionNet	74.84
	Two-stream fusion model: MotionNet +RGB	77.25	32

Table 2. Comparison of experimental results of two models

Method	EER /%	AUC	Detection speed /fps
Method in Ref. [23]	34.9	50.66
Method in Ref. [5]	27.3	65.51	143.50
Method in Ref. [25]	23.6	69.30	0.04
Method in Ref. [21]		75.41
Method in Ref. [24]	18.1	73.68	120.00
Our method	16.8	77.25	32.00

Table 3. Comparisons of the experimental results between the proposed method and other methods

Xianbin Yang, Jianwu Dang, Song Wang, Yangping Wang. Anomaly Event Detection Based on Two-Stream Network and Multi-instance Learning[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2015006

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information