Performance Analysis of Metric Threshold in SURF for Object Tracking

Abstrak

Abstract—Object tracking has a lot of progress every year and gives something new, so the trackers and method itself is getting better. Many researchers are engaged in one of the ï¬elds of computer vision to provide good beneï¬ts for human life in the ï¬eld of Internet of Things. Feature extraction is needed in object tracking processes. One of the scale and rotationinvariant local feature extraction methods, namely Speeded-Up Robust Feature (SURF). In implementing it on object tracking, SURF will extract features from two frames and match them. Then, the Random Sample Consensus (RANSAC) as a rejection correspondence using an inlier on the features obtained and then performs the estimation transformation which is applied to the input match the reference image. In this paper, we analyze one of the parameters of SURF, namely Metric Threshold as a parameter that determines the strongest feature threshold. From the evaluation results, it was found that the default parameters gave non-optimal results.