A Deep Dive into TENSOR’s Technological Innovations: Gait Biometrics

Introduction

TENSOR is a project aimed at enhancing the capabilities of law enforcement agencies in the field of security and law enforcement. Its mission is to provide advanced systems for efficiently and effectively storing and sharing biometric information, ensuring seamless interoperability for various EU LEAs.

Gait biometrics are important for enhancing security measures because they offer a non-intrusive method of identifying individuals based on their manner of walking. This can be particularly useful in situations where other, more robust biometric data, such as fingerprints or facial features, are not available or reliable due to poor viewing angles or camera resolution. Gait biometrics can provide an additional layer of verification, reducing the likelihood of false matches and improving the accuracy of person re-identification across a network of CCTV cameras. Consequently, the TENSOR project incorporates gait biometrics to fully explore its potential value for security enhancement in diverse situations.

Understanding Gait Biometrics

Gait biometrics refer to the unique manner in which an individual walks. Unlike other forms of biometric authentication, gait biometrics are non-intrusive and can be captured at a distance without the subject’s cooperation. This makes them particularly valuable in surveillance and security contexts.

TENSOR utilizes gait biometrics by incorporating advanced technologies for human detection, pose estimation and gait description from video sequences. The extracted gait features are then analyzed and matched against a database to identify individuals.  The main distinctive features are coming from various pre-trained Deep Learning based models, which use sequences of skeletons or silhouettes as an input. Less robust types of gait features analyzed include stride length, cadence, and joint angles, among others.

TENSOR’s Approach to Gait Biometrics

The key components of TENSOR’s gait biometrics technology include human detection and pose estimation from video sequences, a deep learning model for gait biometrics extraction, and components for extracting other gait characteristics. These components work together to capture and analyze individual gait patterns, which are then stored in a database and used for matching and identification.

TENSOR collects and analyzes gait data through video footage, such as CCTV footage of suspects. The data is encrypted and processed through an extraction model, which matches it against a pre-registered gait database to find the best matches for identification. The machine learning algorithms used in TENSOR’s technology include state-of-the-art methods for person detection and pose estimation, a reliable and real-time tracker, and a state-of-the-art gait recognition model.

Addressing Challenges and Ensuring Security

The main challenges in visual gait biometrics include the novelty of this biometric, relatively scarce usage in real applications, and variability in gait patterns due to factors such as injuries, fatigue, or environmental conditions. During the first year of the project, extensive research was conducted to benchmark state-of-the-art gait detection algorithms on real data, revealing a significant gap in gait recognition accuracy between laboratory datasets and real CCTV footage data.

To ensure the security and privacy of gait biometric data, TENSOR implements a biometric data protection mechanism that enables the revocability of biometric templates. This ensures that biometric data can be securely stored and shared among law enforcement agencies in a privacy-preserving manner. Additionally, TENSOR is developing a European Biometric Data Space to facilitate the secure and automated exchange of biometric intelligence among security practitioners. Currently, there is no specific mechanism in place to detect and prevent impersonation or fraudulent attempts to mimic gait patterns within the TENSOR project.

Real-World Applications and Impact

Real-world scenarios remain challenging for visual gait biometrics on a standalone basis. However, in combination with other biometrics, or in scenarios with a limited number of suspects, gait biometrics can provide an additional identification layer beyond traditional methods.

The technology can significantly aid law enforcement agencies and forensic investigations. For instance, it can help identify suspects from surveillance footage where facial recognition is not possible due to poor video quality or obstructions like face masks. Moreover, gait biometrics can verify the identity of individuals in a non-intrusive manner, which is crucial in sensitive environments.

Consortium partners contribute significantly to the development and implementation of TENSOR’s gait biometrics solution by participating in review meetings and providing important feedback for technology development and use-case scenarios. The gait extraction module will be integrated into the Tensor platform developed through the collaborative efforts of all partners.

The expected outcomes of TENSOR’s innovations in gait biometrics include enhanced security measures in public spaces, airports, or border crossings, and improved capabilities for law enforcement agencies and forensic investigations.

Conclusion

In conclusion, the use of gait biometrics in the TENSOR project represents a significant stride forward in the field of security and law enforcement. By leveraging unique walking patterns for identification, TENSOR enhances the capabilities of law enforcement agencies to monitor and secure public spaces, manage access control, and conduct forensic investigations. The non-intrusive nature of gait biometrics, coupled with the ability to capture data at a distance, positions it as a valuable addition to the existing suite of biometric tools.

Looking ahead, advancements in gait biometrics are expected to focus on improving the precision and reliability of gait recognition algorithms, accommodating a variety of conditions and individual behaviors. The incorporation of cutting-edge machine learning and artificial intellingence tools will be crucial in refining these technologies, facilitating more complex analysis and interpretation of gait data. However, the scarcity of datasets collected within the EU and evolving legislative landscapes remain challenges for gait recognition research in the coming years.