On De-Interlacing and Sub-Pixel Precision Tracking

Abstract

Video cameras with interlaced scan sensors find applications in a variety of tasks such as object tracking due to their lower overhead in terms of memory and the higher sensitivity in comparison to their counterparts that employ progressive scan sensors. Such cameras, however, suffer from noticeable interlacing artefacts that need to be corrected with appropriate de-interlacing methods before the target in the video can be accurately tracked. Despite this, the effect of de-interlacing methods on the object tracking accuracy has not yet been widely studied. In this work, the first comprehensive comparison of different de-interlacing methods is carried out in the context human computer interaction studies where precise finger tracking is required. Furthermore, we propose a semiautomatic sub-pixel annotation scheme to create precise ground truth for fingertip location, allowing the analysis of the impact of de-interlacing filters on tracking at sub-pixel level. The experimental part of the work showed that the de-interlacing filter by Pixop outperformed other filters that were evaluated. Moreover, the plausible benefits of sub-pixel precise tracking over pixel precise tracking in trajectory analysis were demonstrated.