End-to-End Deep One-Class Learning for Anomaly Detection in UAV Video Stream.

Slim Hamdi,Samir Bouindour,Mohamed Abid,Tian Wang,Hichem Snoussi

doi:10.3390/jimaging7050090

Abstract

In recent years, the use of drones for surveillance tasks has been on the rise worldwide. However, in the context of anomaly detection, only normal events are available for the learning process. Therefore, the implementation of a generative learning method in an unsupervised mode to solve this problem becomes fundamental. In this context, we propose a new end-to-end architecture capable of generating optical flow images from original UAV images and extracting compact spatio-temporal characteristics for anomaly detection purposes. It is designed with a custom loss function as a sum of three terms, the reconstruction loss (), the generation loss () and the compactness loss () to ensure an efficient classification of the “deep-one” class. In addition, we propose to minimize the effect of UAV motion in video processing by applying background subtraction on optical flow images. We tested our method on very complex datasets called the mini-drone video dataset, and obtained results surpassing existing techniques’ performances with an AUC of 85.3.

Highlights

The use of drones is booming around the world with a large variety of potential applications: wireless acoustic networking for amateur drone surveillance [1], updating of UAV networking using the software-defined radios (SDR) and software-defined networking (SDN) [2], the multi-agent reinforcement learning (MARL) framework [3] and malicious Wi-Fi hotspots detection [4]
Deep Learning (DL) is a sub-domain of Machine Learning (ML), it aims to learn high-level abstractions in data using multi-level architectures
The originality of our work is to propose a new architecture bringing together the advantages of both generative and deep one-class models for anomaly detection purposes in a UAV video footage

Summary

Introduction

The use of drones is booming around the world with a large variety of potential applications: wireless acoustic networking for amateur drone surveillance [1], updating of UAV networking using the software-defined radios (SDR) and software-defined networking (SDN) [2], the multi-agent reinforcement learning (MARL) framework [3] and malicious Wi-Fi hotspots detection [4]. The use of the UAV camera has become very important in the field of detecting abnormal behaviour in video footage This importance stems from the fact that can a UAV monitor large and dangerous areas, but it is cost-effective and can replace an entire installation of fixed cameras [5]. A possible solution to this problem would be the use of intelligent video surveillance systems These systems must be capable of analysing and modelling the normal behaviour of a monitored scene and detecting any abnormal behaviour that could represent a security risk. Considerable technological advances in the fields of machine learning and computer vision have made it possible to process CCTV systems. There are many barriers to the creation of such databases—for example, we can cite the following:

Objectives

Methods

Results

Conclusion