Juan Adarve (PhD Graduate from ANU) and Rob Mahony have created an optical flow algorithm for high-speed applications. They designed the algorithm from stratch with the motivation of being able to process the image stream of a high-speed, high-resolution camera (1Mpix at 200+ Hz) online, and provide a useful visual cue such as optical flow to the robot controller.
They succeeded at this by formulating the optical flow problem as a filtering problem. Rather than computing fully dense flow fields between two frames, which is the common approach in Computer Vision, their algorithm incrementally builds the flow fields by incorporating new image data from the camera. At the heart of the algorithm is a partial differential equation used to predict the future value of flow at each pixel. This prediction is then updated with the new image obtained with the camera, improving the previous estimation.
To showcase their method, they recorded high-speed video while driving a car around the ANU campus, using a Basler USB 3.0 camera to record 1016×544 300 Hz video. The results can be seen in this YouTube video below.
Australian Centre for Robotic Vision
2 George Street Brisbane, 4001
+61 7 3138 7549