OpenCV3.0 has a tracking API involving the implementation of various single-object tracking algorithms. These tracking algorithms include BOOSTING, MIL, KCF, TLD, MEDIANFLOW, MOSSE, and CSRT. As long as we observe the current position of the object, we know its motion, that is, the parameters of the object motion model. Furthermore, based on the location, velocity, and direction of the object in the prior frames, we can predict the new position of the object based on its current motion model. But we require more data. We know what the object looks like in each previous frame. In more detail, we can build an appearance model that simulates the object. This visual model can be employed to search for a small spatial area near the location predicted by the motion model to more precisely predict the position of the object. The motion model predicts the rough location of the object. This accurate calculation is provided for a more precise assessment based on the model. In this study, we first characterized these trackers and then simulated the performance of them using Raspberry Pi hardware. Plus, with the outcomes of the simulations, we compared these trackers.
How to cite: Darvishi H. Comparison of API trackers in OPENCV using Raspberry Pi hardware. SPEC J ELECTRON COMPUT SCI 2020;6(2):1-9