Finding someone in a surveillance video could soon be as easy as Googling them. Descriptions of people of interest, such as a suspect or a missing person, are normally given in terms of their height, gender or clothing. But using this information to find a short woman wearing a red jacket in a video, say, often requires scanning hours of footage manually, which is no easy task. But a new search tool can do it automatically.
The system was created by researchers at Ahmedabad University and Lalbhai Dalpatbhai听College of Engineering in Gujarat, India,听and it uses a form of artificial intelligence algorithm called听machine-learning听to automatically match individuals in videos to their descriptions.
Previous tools let you select someone in one frame of a video and search for other frames in which he or she also appear, but searching using a description is more useful when an eye witness report is all you have to go on. Using height, gender and clothing also lets you find people who are trying to hide their face.
Advertisement
The team trained the software on 110 different videos of people shot from different angles and distances. Given a new video and a physical description, the system looks for people in each frame and calculates their height.
If several people are the same height, it narrows down the selection using the colour of their clothing and whether they are a man or woman. In tests, the software found 28 out of 41 people.
The team plans to improve the accuracy by adding the ability to distinguish between people鈥檚 build, pose and style of clothing. 鈥淭he work will be a valuable contribution to the surveillance systems of the future,鈥� says team member Vandit Gajjar at Ahmedabad University in India.
The听group will present its work at the International Conference on Advanced Video and Signal-based Surveillance in Auckland, New Zealand next month.
Reference:听
