Color theory does not answer this question. However, color science and cognitive psychology might give you a hint.
A human eye is mostly sensitive to brightness, and color second. For example TV does not traditionally color information to all pixels and yet you seem to understand the image fine. Simply the primary edge fining mechanism of the eye (yes, this processing happens structurally on the retina itself) is for most parts only concerned with the brightness of the image. The data in your image is based on edge finding. Therefore, this predicts that there is likely no one color that you can choose for the task.
So contrast, is in simplest form, largely a function of turning the image gray scale and looking at it without color. Therefore simplest approach is to make the arrow out of 2 contrasting colors as others have done. (this is also the reason pS uses marching ants)
Another approach is to blur the data set below. Invert and change brightness of the inverted value slightly and allow that to modify your arrows luminance. Or use a swtooth curves. Many applications use such methods for making selection stand out (like PS does for font selection).