Software to detect objects inside videos
Researchers at the University of Missouri (MU) are developing software that would enable computers to perform video analysis tasks, such as alerting emergency services if a video surveillance camera detects a person falling and not getting up. The software could also be used to search inside videos and look for certain objects, such as basketballs or footballs, hence reducing the time taken to locate a certain game or scene.
“The goal of our research is to improve how computers interpret the content of a video and how to identify it,” said Tony Han, electrical and computer engineering professor in MU’s College of Engineering. “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.”
The researchers believe that human detection software could also be applied to assisted driving. For example, if the software detected a pedestrian in line with the moving vehicle it could make the car stop immediately to avoid an accident. Computer detection might also improve care for older adults living alone. In homes with access to video surveillance, if an older adult fell suddenly, computer detection software could detect the fall and alert medical professionals.
“My students and I are working on algorithms for automatic object detection, but these are very difficult to perfect,” Han said. “We’re trying to find a way to create reliable detection algorithms, but it takes a lot of time to test them. We have manually labeled more than 3,000 images with object locations and have used them to test our algorithms.”
Earlier this year Han and his students from MU attended the PASCAL grand challenge in object detection, where they competed in 20 detection categories against researchers from all over the world. In their first time competing, they won first place in detection for potted plants and chairs and second place in detection for humans, cars, horses and bikes.
Han’s research has been published in numerous publications, such as the IEEE Conference on Computer Vision and the Second IEEE Workshop on CVPR for Human Communicative Behavior Analysis.