Real-Time Anomaly Detection and Localization in Crowded Scenes

TL;DR

This paper introduces a real-time, efficient method for detecting and localizing anomalies in crowded scenes using local and global video descriptors, Gaussian classifiers, and auto-encoders, achieving comparable accuracy to state-of-the-art methods.

Contribution

The proposed approach combines simple Gaussian classifiers with auto-encoder learned features for real-time anomaly detection and localization, improving efficiency while maintaining accuracy.

Findings

Comparable accuracy to state-of-the-art on UCSD ped2 and UMN datasets

More time-efficient than existing methods

Reliable detection and localization of anomalies in real-time

Abstract

In this paper, we propose a method for real-time anomaly detection and localization in crowded scenes. Each video is defined as a set of non-overlapping cubic patches, and is described using two local and global descriptors. These descriptors capture the video properties from different aspects. By incorporating simple and cost-effective Gaussian classifiers, we can distinguish normal activities and anomalies in videos. The local and global features are based on structure similarity between adjacent patches and the features learned in an unsupervised way, using a sparse auto- encoder. Experimental results show that our algorithm is comparable to a state-of-the-art procedure on UCSD ped2 and UMN benchmarks, but even more time-efficient. The experiments confirm that our system can reliably detect and localize anomalies as soon as they happen in a video.