Optical flow estimation using local features

The computation of optical flow by the differential method imposes additional constraints to the one already imposed in the derivation of the optical flow equation. Consequently, the computation of optical flow using differential methods is computationally expensive especially for devices such as mobile phones, which have low processing power. In this work, we propose an optical flow computation method based on local features called the nearest flow. Our nearest flow method works by estimating the distance ratio of two nearest features to find the best match for a feature point. To improve the quality of the sparsely generated flow vectors, we apply the random sampling consensus method to remove false flows that may arise as a result of noise and other imagery conditions. We compare the performance of our nearest flow method with that of Gunner Farneback's and the local differential method of Lucas and Kanade by evaluating the average angular error for each method in the computation of optical flow. The results obtained show that our nearest flow method is faster and more accurate than Gunner Farneback's method and it is almost at the same level of performance as the Lucas and Kanade method.