Abstract
Human Activity Recognition (HAR) has become increasingly crucial in several applications, ranging from motion-driven virtual games to automated video surveillance systems. In these applications, sensors such as smart phone cameras, web cameras or CCTV cameras are used for detecting and tracking physical activities of users. Inevitably, spoof detection in HAR is essential to prevent anomalies and false alarms. To this end, we propose a deep learning based approach that can be used to detect spoofing in various fields such as border control, institutional security and public safety by surveillance cameras. Specifically, in this work, we address the problem of detecting spoofing occurring from video replay attacks, which is more common in such applications. We present a new database containing several videos of users juggling a football, captured under different lighting conditions and using different display and capture devices. We train our models using this database and the proposed system is capable of running in parallel with the HAR algorithms in real-time. Our experimental results show that our approach precisely detects video replay spoofing attacks and generalizes well, even to other applications such as spoof detection in face biometric authentication. Results show that our approach is effective even under resizing and compression artifacts that are common in HAR applications using remote server connections.
Highlights
With the availability of ubiquitous computers and smart wearable sensors, HumanActivity Recognition (HAR) has become one of the popular research topics in recent years.Human Activity Recognition (HAR) algorithms aim to present information on simple or complex physical activities of humans
Within a considered clip, we run our models on three frames that are 15 frames apart and combine these predictions to obtain a single prediction for this clip
Given a video stream, in the case of Single Frame Model (SF) and Ensemble Multi-Stream Model (EM) models, the first prediction is made after 31 frames and, thereafter, predictions are made every
Summary
HAR algorithms aim to present information on simple or complex physical activities of humans. These algorithms take data from various sensors as input and use machine learning and computer vision techniques to extract information on human activities. In the context of automatic HAR, spoofing attack represents a situation when a person misleads the activity recognition algorithm by providing disguised or copied visual data and the algorithm reports these fake data as a successfully performed action. There is a great demand for intelligent video surveillance in places such as, but not limited to, large waiting rooms and campuses to recognize simple or complex motion patterns and derive high-level subjective descriptions of actions and interactions among subjects and/or objects. There is a need for detection of spoofing attacks of this kind before making a final decision on the recognized action
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have