Efficient Scale Estimation Methods using Lightweight Deep Convolutional Neural Networks for Visual Tracking

In recent years, visual tracking methods that are based on discriminative\ncorrelation filters (DCF) have been very promising. However, most of these\nmethods suffer from a lack of robust scale estimation skills. Although a wide\nrange of recent DCF-based methods exploit the features that are extracted from\ndeep convolutional neural networks (CNNs) in their translation model, the scale\nof the visual target is still estimated by hand-crafted features. Whereas the\nexploitation of CNNs imposes a high computational burden, this paper exploits\npre-trained lightweight CNNs models to propose two efficient scale estimation\nmethods, which not only improve the visual tracking performance but also\nprovide acceptable tracking speeds. The proposed methods are formulated based\non either holistic or region representation of convolutional feature maps to\nefficiently integrate into DCF formulations to learn a robust scale model in\nthe frequency domain. Moreover, against the conventional scale estimation\nmethods with iterative feature extraction of different target regions, the\nproposed methods exploit proposed one-pass feature extraction processes that\nsignificantly improve the computational efficiency. Comprehensive experimental\nresults on the OTB-50, OTB-100, TC-128 and VOT-2018 visual tracking datasets\ndemonstrate that the proposed visual tracking methods outperform the\nstate-of-the-art methods, effectively.\n