{"id":882,"date":"2018-01-30T10:45:28","date_gmt":"2018-01-30T10:45:28","guid":{"rendered":"http:\/\/www.mub.eps.manchester.ac.uk\/in-abstract\/?p=882"},"modified":"2018-04-20T09:56:07","modified_gmt":"2018-04-20T08:56:07","slug":"analysis-of-spatio-temporal-representations-for-robust-footstep-recognition-with-deep-residual-neural-networks","status":"publish","type":"post","link":"https:\/\/www.mub.eps.manchester.ac.uk\/in-abstract\/analysis-of-spatio-temporal-representations-for-robust-footstep-recognition-with-deep-residual-neural-networks\/","title":{"rendered":"Analysis of Spatio-temporal Representations for Robust Footstep Recognition with Deep Residual Neural Networks"},"content":{"rendered":"

Learning the way people walk<\/strong><\/p>\n

Human gait is a valuable biometric and various vision systems are currently used to identify individuals by the way they walk. However, the utility of the data collected by cameras depends on the quality of visual access, strongly affected by various parameters of the surrounding scene.
\nResearchers at the University of Manchester have taken a different approach, focussing on non-intrusive gait recognition by monitoring the force exerted on the floor during a footstep cycle. This is challenging, because the features which distinguish the subtle variations from person to person are difficult to define. Instead, we use the raw data acquired from known users for training artificial neural network models, to identify someone who is unknown to the system, using gait features automatically extracted by the model.
\nIn collaboration with the University of Madrid, we have validated this approach in the largest to date public footstep database, containing 20,000 footstep signals from more than 120 users, in 3 critical data-driven security scenarios: airport security checkpoints (smallest training set), workspace environments (medium training set) and home environments (largest training set). In the latter case we achieve 0.7% error, when the rates of false rejection and false acceptance are equal, which is 3.7 times better than the state-of-the-art. The methodology is now being developed to address the healthcare problem of markers for cognitive decline and onset of mental illness, by using raw footstep data from a wide-area floor sensor deployable in smart dwellings.<\/p>\n

<\/p>\n