(+86) 18081079313; (+86)28-86256789
Title: Abnormal Behaviors Detection in Surveillance Video
Background: Abnormal behaviors in public space can be potentially hazardous. For example, crowd congestion may cause a stampede if the density is too high, and panic dispersing often implies an emergent situation. Ordinary approach to handle the surveillance video from the publicly installed cameras is to arrange human operators by monitoring screens. However, its performance is greatly impacted by the fatigue and distraction. On the other hand, the information distribution of surveillance data is often very sparse. To reveal the crucial information, analyst must inspect the entire video which is extremely time consuming. Therefore, a computer vision-based approach to effectively detect the anomalies can be exploited to address the above-mentioned issues.
Data: (1) Bench-marking datasets: KTH, UCF and Weizmann
(2) Provided dataset: 440 clips of videos, with 9 categorizes of
behaviors including sabotaging public properties, shooting, holding
blades, fist fights, robbery, flee, thievery, assault and normal
behaviors.
Task: To devise an approach to recognize the
individual/crowd behaviors with either conventional-based or deep
learning-based techniques. The proposed approach should provide
satisfactory performance on behavior classification or anomaly
detection, which can be utilized for the prediction of potentially
hazards and video abstraction. The approach should have a good
precision on the bench-marking datasets, and show promising
effectiveness on the provided behavior dataset to prove its
robustness on variable data qualities.
Requests:
Participants can be either university students, graduated researchers or from industries.
One team can include 1-3 team members, and 1-2 advisers.
Participants need to provide source codes, trained model, algorithm description and testing results.
The performance of the proposed approach is evaluated by the accuracy on both benchmarking and provided datasets.
Milestones of the challenge:
Nov. 1. 2021 – Nov. 20. 2021 Registering for the competition
Nov. 9. 2021 – Feb. 21. 2022 Team work
Feb. 22. 2022 Submission
Feb. 22. 2022 – March. 10. 2022 Testing, performance evaluation, ranking
March. 10. 2022 Winner notification, Competition close
Organizer:
Xi'an University of Posts and Telecommunications
Data provided by:
Center for Image and Information Processing, Xi'an University of Posts and Telecommunications, Xi'an, China, 710121.
Crime Scene Investigation Unit, Department of Public Security, Shaanxi, China.
For challenge registration and any inquiry, please email us at: haoyu@xupt.edu.cn.