Here at CAST, our objective is always to advance existing technology with the requirements of our clients in mind. To this end, we seek to work with those who have exciting new ideas and fresh perspectives. As part of our association with the University of Otago, we have this year worked closely with an intern student, Brett Jones, and a PhD graduate, Tum Thongpaibool, on an object tracking solution that could be a game changer for monitoring the location and condition of road assets – among a number of other applications.
High Quality Object Detection
A neural-network object detector was trained to automatically recognize several classes of road signs from video footage. The example below shows the detector in action, identifying signs, frame by frame, with a high degree of accuracy in New Zealand highway footage. The numbers displayed indicate the degree of confidence in the quality of the detection. Of note, the detector can be readily retrained to other object types and reapplied to any video footage – thus significantly speeding up essential road surveys that count and assess road signage, guard rails, distance markers and more.
While these results are impressive, framewise object detection is only the first step toward practical application.
Object tracking
To be useful for most purposes, accurate object detection needs to be combined with accurate object tracking. That is, recognizing which objects detected in one video frame are the identical objects in subsequent video frames. This challenge is made harder with a moving camera and objects that become temporarily occluded from view when they pass behind other objects. Tracking is essential for accurate counts of unique objects and other performance and classification metrics. For example, accurate tracking would make it possible to complete full road stocktakes that account for signage condition, missing road marker posts, cleaning requirements and where replacements are needed. The technology could build a comprehensive database of the road system, creating huge savings and improving safety.
We are there – but also, not quite
We don’t want to get ahead of ourselves – some of the features needed for tracking are still a work in progress. But we can see huge potential from Brett and Tum’s work, and we expect that the full capabilities will be revealed in just a few short months.
As you can see from our video, a world where computer vision solutions are truly possible, has arrived. Not only can computers track objects across video, but they can apply metrics such as counts in real time. With CASTs computer vision solutions, high risk and time-consuming jobs can now be automated to reduce risk and injury, redirecting essential labour resources to more meaningful areas of your business.
Want to hear more about object tracking solutions and how they could apply to your business? Give us a call.