A good reference point is a high-contrast pattern that has good definition both vertically and horizontally, which allows for perfect registration in both directions. Preferably, the reference point should not change much over time.
To choose a good reference point, play the footage several times to become familiar with the material. Ideally, you should try to find a pattern that is present in every frame. However, this is not always possible, and you may have to track two different features, such as object that disappears behind another one, or an object that moves out of the frame. The reference frame (the frame on which you place the reference box) should be the frame where the reference point is most representative in terms of shape, size, and rotation. Generally, you position the reference box over the reference point in the first frame of the sequence, but you could also place it on the last frame and analyze backwards. Another alternative is to place it on a frame in the middle of the footage and analyze from that frame backward to frame 1, and then analyze from that frame forward to the end of the footage.