Prototype-based budget maintenance for tracking in depth videos

Abstract

The use of conventional video tracking based on color or gray-level videos often raises concerns about the privacy of the tracked targets. To alleviate this issue, this paper presents a novel tracker that operates solely from depth data. The proposed tracker is designed as an extension of the popular Struck algorithm which leverages the effective framework of structural SVM. The main contributions of our paper are: i) a dedicated depth feature based on local depth patterns, ii) a heuristic for handling view occlusions in depth frames, and iii) a technique for keeping the number of the support vectors within a given “budget” so as to limit computational costs. Experimental results over the challenging Princeton Tracking Benchmark (PTB) dataset report a remarkable accuracy compared to the original Struck tracker and other state-of-the-art trackers using depth and RGB data.