coordinates The coordinates can’t be defined when the matching points are clinging to moving objects. This means that the filtration isn’t working well enough. The coordinates can’t be defined also when camera rotation angle is more than 90°. As a solution to the first problem we now consider applying video motion segmentation to distinguish static points from motion points (taking into consideration the nature of movement). As a solution to the second problem we see the transfer to the cylindrical coordinate system.
To find the homography you need at least 4 matching points. But in some cases the 4 points can’t be found, and the homography matrices are Н=None. In a current algorithm version we process such cases this way: if the argument none_H_processing is set for True we consider the matrix of the previous frame matches the matrix for the current frame (Hk=Hk-1). If set for False, then H=[[1,0,0][0,1,0][0,0,1]], meaning that there were no movement in the frame. It is necessary to think over better handling of such situations.
There’s an inaccuracy in the coordinates. Poorly obtained homography matrix distorts the results of coordinate recalculation. The reasons for that are: Poor filtration. If the points catch on a motion object, then the homography matrix will describe not only the camera movement, but also the independent movement of objects (for example, a person's walking). Using the built-in findHomography () function of the OpenCV module. This function already assumes there is an error in the calculation of the