dorazaria / spacetracker Goto Github PK

Using background subtraction algorithm of Gaussian Mixture-based Background Segmentation, 
with a very short history it can detect moving objects.
It generates a BnW frame, the whiter, the faster.
We used another algorithm to detect the whiter object and then analyse it for correctness.
After detecting, we used some limited-data-structures that contains:
1) 'stat' - the position of an object in each frame.
           If the new position of the detected object is around the mean probability of 'stat' list
           then it probably an object that moves logically.
2) 'avg_contours' - the number of total objects we can get in each frame.
                   By probabilistic calculation of variance and mean this list can decide 
                   if the picture is 'loud' by many objects (it happen mostly near the ground).
                   So if the variance and mean is big, the position of the detected object
                   can't go far from the last position it was because there's too many objects.
                   If the variance and mean is small, we can look for a far position of the object 
                   comparing to its last position.
          
When the program detect an object, the user can press specific keys to control the state.
Using smart tracker algorithm of CSRT -(Discriminative Correlation Filter with Channel 
and Spatial Reliability) which is specializing in fast-moving objects.
          
The user can press:
   - 'space' to track a detected object.
   - 'c' to cancel the tracking and move to detection scenario.
   - 'z' to zoom-in to the object.
               
The transition from sky to ground (and vice-versa) is safe while we track an object even though the two  
algorithms are completely different, it is a safe-change between algorithms without losing the object.

Given a color by the user (RED, YELLOW or ORANGE).
The algorithm detects the object by color.
Using color range of the given color, it detect the most 
stronger color and relevant-sized object.
           
Using HSV color picture we search in each frame the correct color we wish to detect
and then using mask we find the right contours (shapes).
           
When the program detect an object, the user can press specific keys to control the state.
Using smart tracker algorithm of CSRT -(Discriminative Correlation Filter with Channel and 
Spatial Reliability) which is specializing in fast-moving objects.

Using MOG2 algorithm we can detect the moving objects.
Then we picked the fastest one (which is presented by the strongest white color).
The detect object is displayed to the screen and the user can chose to track it or not.

This method suggest the size of radius of detecting-search
using the mean of the history list of the number of moving objects we found
in each frame.
If the number is big, then look somewhere near because we can't predict where the object
has lost if there is too many objects around.

If the number is small, then it can search by a big radius for lost the object.

:return an integer value (the radius).

This method decides using color range if the frame is at ground or sky mode.
        
:return True if it's sky mode.