Algorithm A2. Trajectory synthesis procedure

Input:  Video sequences (1-4)

Output :  List of tracking rectangles coordinates of the most visible position from all views/      frame file; frame number of the analyzed position and the movement attribute   (moving, freezing).

01.  Choose the camera (view) set to be included in the project; Generate the list results: one list for each                 project camera; one list for fused trajectory; For each project video frame is allocated an entry list include      ing: camera number; localization coordinates and movement attributes.

02.  Establish the tracking mode: parallel (two to four camera tasks) or sequential (for debug                                       purpose);

03. for each project camera do

04.  set the polygon of useful view of the entire camera scene;

05.  set the areas    of doors scene through panda or zoo technician could enter in or quit the     scene;

06.   set the areas where could not be considered asleep in case of lack of movement features     for the current camera;

07.   set the areas where the movement features should be erased because are generated by     the time and frame counters or by the water waves for the views including water ponds     (mainly the fourth view);

08. end for

09. for each frame, denoted iFrame, of the current video sequence (start_frame to    end_frame) do

10.  if iFrame–start_Frame < 3     compute the gray level average (denoted gray_level_average_current) for each view;

11.   if at least one gray_level_average_current is too low or too high (the image is entire      black or white)

12.     start_Frame = iFrame;

13.    continue; end if

14.     fill the list gray_level_average[3] of each view:

task(i).gray_level_average[2] = task(i). gray_level_average[1]

task(i).gray_level_average[1] = task(i) .gray_level_average[0]

task(i).gray_level_average[0] = task(i). gray_level_average_current;

$i=1,m; \mathit{m}\in \left\{2,3,4\right\}$

15.    continue; end if

16.  if parallel _mode launch tracking task (iFrame, camera_nmb, start_frame) of the      all cameras in the project;

15.  else

16.   for each camera in project execute tracking task (iFrame, camera_nmb, start_frame);     end for, end if

17.   if at least one of the tasks returns a value related to zoo technician presence in the scene       do not consider the current frame in the fusion process; end if

18.  if a task returns non valid coordinates, others than those signalizing the technician       presence do not consider it in the fusion process; end if

19.  if only one task returns valid position retain its view as most suitable view; end if

20.  if at least two tasks return valid positions retain most suitable view considering area      localization weight and the previous view in the fused views chain;

21.    if the previous selection is near the border of a full view area, neighbor of another full       view area, and the movement speed is fast (lot of movement features) do not      switch views

else

22.     if the previous selection is near the fading view border of a full view area and at least         one of the others available views is in a full view area consider this one instead the         current view; end if (lines 22, 21, 20)

23.  end for;