Video Analytics

Analysing Video Data for Retail Space intelligence and Surveillance

 

CVRG is applying advanced computer vision and pattern recognition techniques to address intelligent space design and surveillance applications.

An emerging need is to automatically monitor space in an attempt to detect events and object-of-interest. Monitoring by human operators is a tedious task and requires a large number of personnel, resulting in high ongoing costs and questionable reliability as the attention span of human operators decreases rapidly when performing such mundane tasks. A solution may be found in advanced computer surveillance systems to monitor all video feeds and deliver alerts to human responders for triage — a well-designed computer system is never caught “off guard”.

One key outcome of this project is a fast and robust multi-camera based face matching and search framework, which can be used to detect persons of interest. Unlike many existing solutions, it is designed to work with low-resolution images of ‘uncooperative’ subjects (people who are not posing for the camera). It is robust to variations in environmental conditions, quality and pose. Additionally, it has a comparatively small computational footprint and can be parallelised, thus making it easily scalable.

As it is often difficult to acquire good face images, this technology is being combined with our research in person tracking to achieve a better estimate of identity combining several biometric techniques over time.

Our work aims to provide better security through non-intrusive intelligent surveillance. This tool can assist in forensic examination of video after an incident or potentially proactively alert the authorities at the outset. The techniques can also be applied outside of security and surveillance, such as facial recognition for photo-tagging in consumer products.