The question Is it possible to represent an orbit with just a periapsis position and 2 angular velocities? got me thinking, suppose I take a telescope and took 3 measurements of ISS angular position and their time stamps can I do OD? It occurred to me that 2 measurements are surely not enough if orbit is eccentric, then next question is, then does 3 measurement suffice? If so can it be done analytically ? If not then what could be an easy way to do numerically ?
As the answer to that question states:
In general, an epoch time and six scalar values that completely represent the state at the epoch time are needed.
The six values you can choose as you like, as long as they are independent variables. For example the position in x,y,z and the velocities in x,y,z directions would suffice. You can choose polar coordinates (r,$\theta$,$\phi$) instead.
With each of your measurements you get two angular positions. From the difference of them you can determine the angular velocities. This gives you already four values out of six. So far you don't have a measure of the distance of the observed object (r in polar coordinates). You can get this by doing another measurement, e.g. one orbit later. This gives you the period (trivial) and the orientation of the orbit (by some further calculations). So, you have six independent pieces of information, enough to exactly specify the orbit (in combination with your time measurement).