Invited seminar in Tübingen
In March 2018, Dr. Omar Costilla Reyes gave an invited seminar at the Max Planck Institute for Intelligent Systems, and the talk was listed on the website of the institute’s Perceiving Systems department for its Tübingen campus. He gave the seminar while completing his doctorate at the University of Manchester, after the underlying paper had appeared in IEEE Transactions on Pattern Analysis and Machine Intelligence.
Footsteps as a biometric signal
The talk framed footsteps as a behavioral pattern that can anchor a biometric system, in contrast to traditional security based on passwords or access cards. Biometric recognition identifies or verifies a person from physical or behavioral characteristics; here the characteristic is the pressure pattern of a footstep recorded by a floor sensor. He presented raw and processed spatio-temporal footstep representations, meaning pressure maps that also change over time, evaluated with deep models built on a two-stream residual-network architecture: one network reads a single footstep image, the other a sequence of frames, and their class scores are combined, as the figure on this page shows. The listing describes the SFootBD footstep database and reports results for three data-driven security scenarios, from airport checkpoints with the least training data to home environments with the most.
Connection to his doctoral research
The seminar presented the biometric strand of his thesis on pattern recognition from raw spatio-temporal floor-sensor data, which also addressed healthcare applications of gait analysis. In the published paper, one tested configuration reached a low equal error rate, meaning that wrongly accepted impostors and wrongly rejected genuine users were both rare, which supports footsteps as a practical biometric signature that a floor sensor captures without a camera. The same research appears in his talks at the University of Cambridge Computer Laboratory and the 2018 Advances in Data Science Conference, and in the book chapter “Deep learning in gait analysis for security and healthcare.”