ESR photo Research project "Privacy preservation in video-based AAL applications" About the project Visual data exposes a lot of information about individuals appearing on images and videos. Individuals may want to conceal all of this data, but in this case, the remaining information would be useless for the AAL services that build upon it. Therefore, there is a need to establish a trade-off between privacy and intelligibility of the images. This project will advance in a privacy-by-context approach, in which different visualisations are produced depending on the context in which images or videos are captured: Identity, appearance, location, ongoing activity of the subject being monitored; event triggered; identity and access rights of the observer; closeness between observer and monitored subject… While these privacy-by-context approach has been successfully employed using RGB-D cameras, it has been difficult to address privacy preservation using regular RGB cameras either located in the environment (preferably on the ceiling) or worn by the user. Therefore, this project will investigate visualisations methods to conceal visual privacy in applications and services for older and frail people that employ RGB cameras. Start date: April 2021 Expected end date: April 2024 Progress of the project This project focusses on the creation of machine learning algorithms for visual privacy preservation in AAL, and more specifically operating on visuals acquired from omnidirectional cameras located on the ceiling of a room. As there is a lack of datasets containing videos from such cameras to train machine learning models for the task, this project is currently focussed on the creation of a novel dataset that synchronises many different modalities, including visuals recorded from RGB-D (RGB, depth and infrared) side-view cameras, omnidirectional RGB cameras with top view, wearable egocentric cameras, and a wearable device that records various biomarkers of the human body such as wrist motion, heart rate, skin temperature, and galvanic skin response. This dataset will be employed during the remaining of the project for different tasks: human segmentation and human pose estimation from top-view images, and activity recognition. This dataset will also serve for ESR14’s research project. Additionally, an extensive review of the state of the art in visual privacy preservation techniques for active and assisted living has also been carried out and submitted for publication. Alongside this, in a collaboration as a result of the secondment at UA by ESR11, a study of the fairness of common visual privacy preservation algorithms has also been conducted and published. Scientific publications ODIN: An OmniDirectional INdoor dataset capturing Activities of Daily Living from multiple synchronized modalities Siddharth Ravi, Pau Climent-Perez, Théo Morales, Carlo Huesca-Spairani, Kooshan Hashemifard, Francisco Florez-Revuelta ODIN: An OmniDirectional INdoor dataset capturing Activities of Daily Living from multiple synchronized modalities 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Vancouver, BC, Canada, 2023, pp. 6488-6497 Fairly Private: Investigating The Fairness of Visual Privacy Preservation Algorithms Sophie Noiret, Siddharth Ravi, Martin Kampel, Francisco Florez-Revuelta Fairly Private: Investigating The Fairness of Visual Privacy Preservation Algorithms Fourth AAAI Workshop on Privacy-Preserving Artificial Intelligence (PPAI-23), Washington, DC, USA, 2023 On The Nature of Misidentification With Privacy Preserving Algorithms Sophie Noiret, Siddharth Ravi, Martin Kampel, Francisco Florez-Revuelta On The Nature of Misidentification With Privacy Preserving Algorithms In Proceedings of the 15th International Conference on PErvasive Technologies Related to Assistive Environments, pp. 422-424, 2022. A Review on Visual Privacy Preservation Techniques for Active and Assisted Living Siddharth Ravi, Pau Climent-Pérez, Francisco Florez-Revuelta A Review on Visual Privacy Preservation Techniques for Active and Assisted Living arXiv:2112.09422 About the ESR Siddharth holds a master’s degree in Systems and Control (2017), specializing in cognitive robotics from the Delft University of Technology (TU Delft) in The Netherlands. He has since done machine learning research in both industrial and academic settings. His latest stint was at the Norwegian company Q-Free, where he worked on creating novel deep learning-based object detection and segmentation pipelines to solve hard problems related to traffic and transportation. Contact information Siddharth Ravi University of Alicante Department of Computing Technology Ctra. San Vicente del Raspeig, S/N 03690 San Vicente del Raspeig, Spain Email address: siddharth.ravi@ua.es
ODIN: An OmniDirectional INdoor dataset capturing Activities of Daily Living from multiple synchronized modalities Siddharth Ravi, Pau Climent-Perez, Théo Morales, Carlo Huesca-Spairani, Kooshan Hashemifard, Francisco Florez-Revuelta ODIN: An OmniDirectional INdoor dataset capturing Activities of Daily Living from multiple synchronized modalities 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Vancouver, BC, Canada, 2023, pp. 6488-6497
Fairly Private: Investigating The Fairness of Visual Privacy Preservation Algorithms Sophie Noiret, Siddharth Ravi, Martin Kampel, Francisco Florez-Revuelta Fairly Private: Investigating The Fairness of Visual Privacy Preservation Algorithms Fourth AAAI Workshop on Privacy-Preserving Artificial Intelligence (PPAI-23), Washington, DC, USA, 2023
On The Nature of Misidentification With Privacy Preserving Algorithms Sophie Noiret, Siddharth Ravi, Martin Kampel, Francisco Florez-Revuelta On The Nature of Misidentification With Privacy Preserving Algorithms In Proceedings of the 15th International Conference on PErvasive Technologies Related to Assistive Environments, pp. 422-424, 2022.
A Review on Visual Privacy Preservation Techniques for Active and Assisted Living Siddharth Ravi, Pau Climent-Pérez, Francisco Florez-Revuelta A Review on Visual Privacy Preservation Techniques for Active and Assisted Living arXiv:2112.09422