2019/01/28 by Felix Stephenson, Stephenson, Felix, Toby P. Breckon +3
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Vision and Imaging #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1901.09971
openalex publication_date 2019/01/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Modern optical flow methods make use of salient scene feature points detected\nand matched within the scene as a basis for sparse-to-dense optical flow\nestimation. Current feature detectors however either give sparse, non uniform\npoint clouds (resulting in flow inaccuracies) or lack the efficiency for\nframe-rate real-time applications. In this work we use the novel Dense Gradient\nBased Features (DeGraF) as the input to a sparse-to-dense optical flow scheme.\nThis consists of three stages: 1) efficient detection of uniformly distributed\nDense Gradient Based Features (DeGraF); 2) feature tracking via robust local\noptical flow; and 3) edge preserving flow interpolation to recover overall\ndense optical flow. The tunable density and uniformity of DeGraF features yield\nsuperior dense optical flow estimation compared to other popular feature\ndetectors within this three stage pipeline. Furthermore, the comparable speed\nof feature detection also lends itself well to the aim of real-time optical\nflow recovery. Evaluation on established real-world benchmark datasets show\ntest performance in an autonomous vehicle setting where DeGraF-Flow shows\npromising results in terms of accuracy with competitive computational\nefficiency among non-GPU based methods, including a marked increase in speed\nover the conceptually similar EpicFlow approach.\n