Visual Object Tracking: The Initialisation Problem

Abstract

Model initialisation is an important component of object tracking. Tracking algorithms are generally provided with the first frame of a sequence and a bounding box (BB) indicating the location of the object. This BB may contain a large number of background pixels in addition to the object and can lead to parts-based tracking algorithms initialising their object models in background regions of the BB. In this paper, we tackle this as a missing labels problem, marking pixels sufficiently away from the BB as belonging to the background and learning the labels of the unknown pixels. Three techniques, One-Class SVM (OC-SVM), Sampled-Based Background Model (SBBM) (a novel background model based on pixel samples), and Learning Based Digital Matting (LBDM), are adapted to the problem. These are evaluated with leave-one-video-out cross-validation on the VOT2016 tracking benchmark. Our evaluation shows both OC-SVMs and SBBM are capable of providing a good level of segmentation accuracy but are too parameter-dependent to be used in real-world scenarios. We show that LBDM achieves significantly increased performance with parameters selected by cross validation and we show that it is robust to parameter variation.

The Seventh Visual Object Tracking VOT2019 Challenge Results

“…The best of these are locally optimised in a small region around each patch. PBTS uses an alpha mattingbased patch initialisation technique [21] to place patches in regions of the bounding box that most likely contain the object.…”

Initial Matting-Guided Visual Tracking With Siamese Network

“…As described in Section 3, we only add the matting algorithm to the tracking process (experiments on OTB50 are shown in FIGUER 6). It appears that S+Siam is better than the other two methods, and it is worth noting that L+Siam, in which LBDM has the best segmentation of the three [33], does not achieve the best tracking performance, most likely because of over-fitting caused by high segmentation accuracy. Although both are based on superpixel segmentation, SBBM could easily retain more structural information compared with OC-SVM, thus making S+Siam more robust at tracking than O+Siam.…”

“…Usually in tracking, there are both foreground and background in the bounding box given in the initial frame. In order to solve the problem of label uncertainty in the bounding box, [33] proposes three foreground segmentation methods learning based digital matting (LBDM), one-class SVM (OC-SVM) and sampled-based background model (SBBM), the first two of which are based on superpixels. The segmentation results are shown in FIGUER 5.…”