Determination of Optimum Dynamic Threshold for Visual Object Tracker
Asfak Ali, Avra Ghosh, Sheli Sinha Chaudhuri · 2021 International Conference on Automation, Control and Mechatronics for Industry 4.0 (ACMI) · 2021
Mobile object tracking is one of the integral parts of computer vision study. Different algorithms have been developed and open-sourced in computer vision applications, yet, there is huge scope for improvement to fulfill real-time application. Keeping this requirement in mind a new object tracker model has been proposed in this paper which dynamically selects between the Mean-shift algorithm and Unscented Kalman Filter (UKF). The novelty of the proposed method is finding a threshold value (β), which is defined for this selection purpose whose value is computed taking into account the analytical and logical decisions to determine the selection of the above algorithms. The threshold value is calculated dynamically depending upon the environment such as nonlinear motion, non-rigid object deformation, high velocity, blurriness, etc in tracking objects for real-time application and it also helps to minimize the time delay. To improve the robustness of the tracker due to changes in the pose or the light illumination of the frame, this method updates the target model based on the current and previous frames.