With the proliferation of surveillance cameras, society needs means to identify people from the images these cameras provide. Crime-solving websites are replete with imagery of criminals who are often disguised and/or at low resolution; terrorist attacks yield more imagery. We noticed this many years ago and were the first to develop systems that aimed to recognize people by their gait and their style of walking.
This talk will describe some of the earlier approaches and their motivation, together with the recent works on deep learning. More recently we have moved to recognize from human descriptions, consistent with eyewitness statements and the limited spatial and temporal resolution of surveillance imagery, and the chance of disguise. We have shown that human descriptions can be used for recognition and retrieval and formulated ways to make these descriptions more effective. We have so far used descriptions of the face, the body, and the clothing, and our current work shows how the labels can be derived by computer vision and explore the new information available by the interface between semantic description and automated recognition. This talk thus surveys these areas, describing progress in gait and in soft biometrics.
Speaker(s): Prof Nixon,
Virtual: https://events.vtools.ieee.org/m/324472
This event has passed.
