Object Tracking based on Compressive Sensing Using Gabor Filters
Main Article Content
Abstract
The development of efficient and effective appearance models for object tracking is challenging due to factors such as camera angle changes, illumination variations, occlusion, and motion blur. Existing online tracking algorithms often update their models with samples from observations in the current frames. When these adaptive appearance models are data-dependent, there is insufficient data for online algorithms to learn at the beginning. Moreover, online tracking algorithms frequently face the problem of drift. Due to self-learning, there is a probability of adding misallocated samples to the appearance model, which degrades it. In this paper, we propose a simple yet effective tracking algorithm with an appearance model based on feature extraction using a data-independent Gabor filter. The proposed appearance model non-adaptively preserves the structural feature space of the original image of the object. A Gabor filter is designed to effectively extract features for the appearance model when applied to the image. This filter extracts both foreground and background features. The tracking task is formulated as a binary classification problem, utilizing a Support Vector Machine (SVM) classifier with online updates. The proposed tracking algorithm operates in real-time and outperforms other advanced methods in terms of accuracy and efficiency on challenging sequences. The algorithm was tested on six datasets named David, Bolt, Pedestrian, Goat, Cyclist, and Chase. The results from these experiments demonstrate the superiority of this method compared to other approaches.