MBT: A Context Aware Mission Based Tracker

Abstract

Implementing a robust tracker production is a challenging topic because of problematic cases, such as partial occlusion, fully occlusion, deformation, motion blur, fast motion, illumination variation, background clutter and scale variations. On the other hand an object to track has lots of features, such as color, speed, shape, height as well as the context. In order to produce robust object tracking system, use of these features are important. In this paper, the implementation of the Context Aware Mission Based Tracker (MBT) which is a context aware scale adaptive object tracking system based on fast Discriminative Scale Space Tracking (fDSST) is explained. MBT is integrated with color detection, image segmentation, fast move and stabilization mode to introduce context awareness. In addition to the modes, we constructed two different blocks which are velocity control block (VCB) and occlusion control block (OCB) to improve the fDSST and a Kalman filter to increase the location estimation performance. MBT is checked against other object tracing algorithms through the experiments with the Online Tracking Benchmark (OTB50) dataset. Results indicate that tracking performance is significantly improved by our contribution and became the best among the discriminative correlation filter (DCF) based trackers. Although is the ability to quickly calculate the response values of a large number of candidate samples with fast Fourier transform (FFT), making the DCF-based tracker much faster than convolutional neural networks (CNNs) based trackers. Thereby, MBT tracks the target object without CNNs applications.