low-compute object detection. I tried many ways to do this, but the best I found was by using contrast and edge detection. The code tracks the object, forecasts its next frame position based on its velocity, and recalculates for the next frame based on the difference between forecasted and true position.