Recognising people and objects as they move around a city

Professor Aihua Zheng of Anhui University, China visited recently to tell us all about her work to identify the same people and objects in images from different cameras. Imagine a person walking across a city at dusk. The lighting changes from daylight to pools of street lighting. Bright car headlights shine sporadically. Maybe the person puts on a coat, so looks different on camera. Prof Zheng’s talk gave insight into the way that artificial intelligence can be trained to work out if multiple images from different cameras are showing the same person as they move around, under difficult lighting conditions. It helps to use a mixture of infra-red cameras, as well as cameras that work with visible light. Special datasets are used to train the artificial intelligence. One example of the detail that is required is the need to be able to predict the size of the burst of illumination that is created by car headlights. The system can then compensate for the parts of images that are being disguised by this sudden flare of brightness.

Prof Zheng also works on the complementary topic of audio-matching. This is the ability to match voice and face from different video and photos. All this expertise will help as COG-MHEAR researchers work out how to identify speech from one person talking in a room full of conversations and noise.