Système de sécurité biométrique multimodal par imagerie, dédié au contrôle d’accès

Abstract : These thesis works are part of a national project aiming to secure storage and transports of radioactive sources, representing an important security issue. The objective is to design a technical solution addressing the secure of these sources in their storage phase. The proposed solution, allowing authorized staff person authentication, consists in a multimodal biometric security system, based on computer vision and artificial intelligence. The actual industrial biometric legal framework relates to preserve users’ privacy. Therefore, the personal data protection is an important aspect. Using the face, fingerprint and finger veins modalities, the biometric models are individual and only stored on a nominative RFID card. This kind of remote media only have a small user’s memory, thus, the leading research axis is focused on a matching between the algorithms and the processing unit in charge of the computation tasks. The amount of biometric data has then been minimized in order to be stored on the remote media. Manifold leads have been investigated for the face modality, with a comparison of classical Machine Learning algorithms and others from Deep Learning. Various preprocessing has been evaluated to lower the impact of environmental variations on the acquisition. The implemented algorithms dealing with both finger modalities include preprocessing algorithms, one of which are a bank of Gabor filters and a squeletonization. These preprocessing tasks help to detect points of interest. A descriptor locally describes these points and a matching is done then between references descriptors (stored on the remote card) and descriptors from the image acquired during the authentication. The attributes extracted from this matching provide, by a classification, a validation or a rejection of the authentication. For each studied modality, the biometric data stored on the remote media does not exceed 2,6 Ko. Thereafter, a decision fusion from these three modalities gives a global authentication which yields to a better robustness against intruders or spoofing attacks. With a strong industrial aspect, this work put forward a hardware integration of developed methods. Moreover, an algorithm selection ensures real time operation of the authentication task, with respect to the chosen computational unit.
Complete list of metadatas

Cited literature [419 references]  Display  Hide  Download
Contributor : Pierre Bonazza <>
Submitted on : Thursday, July 11, 2019 - 4:50:56 PM
Last modification on : Sunday, July 14, 2019 - 2:13:17 AM


Files produced by the author(s)


  • HAL Id : tel-02180903, version 1


Pierre Bonazza. Système de sécurité biométrique multimodal par imagerie, dédié au contrôle d’accès. Intelligence artificielle [cs.AI]. Université de Bourgogne Franche-Comté, 2019. Français. ⟨tel-02180903⟩



Record views


Files downloads