Object-Detection Software to Enable Search Within Videos

Detection algorithms help computers find humans, or anything else, in YouTube videos or surveillance footage

Imagine running a Google search for basketball videos, and having your computer sift through actual footage of online videos rather than just the text of the descriptions. A new type of software could enable computers to run searches inside videos, and pick out humans and objects alike.

Such detection software could go beyond Internet searches. Intelligent video surveillance might automatically send out alerts if cameras spot an injured or falling person, and car computers could make vehicles stop if a human figure looms ahead on the road. More autonomous drones flying surveillance or rescue missions could also find a use for such software.

“There are lots of possibilities with video-based detection, and it could come at quite a low cost compared to object and human detection using other sensors, such as thermal sensors,” said Tony Han, an electrical and computer engineer at the University of Missouri in Columbia.

Han’s group recently attended the PASCAL grand challenge in object detection, where they won first place for detecting potted plants and chairs, and second place in detecting humans, cars, horses and bikes. The researchers have already labeled more than 3,000 images with object locations to test their algorithms, but continue working to improve the algorithms to provide reliable detection.

We’re still a long way from computers that can pick out LeBron James or Kobe Bryant in video footage, and Waldo remains safe for now in his red and white camo. But it’s an interesting step beyond existing technology, such as an effort by Google to pick out landmarks from your vacation photos.