- Why didn't you consider projecting the coordinates of obstacle (top part) to the fish tank ground plane? Why did you instead chose to project the boundary corners from the ground plan to the obstacle plane?
This question is related to the description contained in Section 3.2 Location of virtual gates created by the obstacles. When we were developing this method, we first tried to project the coordinates of obstacle (top part) to the ground plane which lies submerged under water. Due to refraction of light, the projected coordinates did not align with the obstacle (bottom part). If the obstacles were not submerged under water, there would have been no difference in projecting obstacle plane points to ground plane or vice versa. Therefore, to reduce the impact of mis-alignment caused by refraction of light, we chose to project the four tank boundary corner points from ground plane to obstacle plane instead of projecting several obstacle (top part) coordinates to the ground plane.
- I have a large number of videos. Can I process all of them at once?
Yes. It is possible to re-use the calibration points if all the videos were captured without moving the camera or the fish tank. In other words, if the four corners of the fish tank as remains consistent in all the input videos, the calibration points can be reused as the vanishing point remains consistent. See Section Processing Multiple Videos for more details.
- The List Annotator (LISA) manual annotation tool does not show images even after I correct set the path for video frames?
Web browsers often restrict absolute path (e.g. /data/fish/allframes/video1/f00001.jpg) based access of files stored in the local disk. This is due to security reasons. Here is a quick method to overcome this restriction. Let us assume that all the video frames are extracted and stored in the /data/fish/allframes/ folder with frames corresponding to each video stored under a subfolder name same as the video filename (e.g. /data/fish/allframes/video1/f0001.jpg, /data/fish/allframes/video2/f0001.jpg, ...). Open the LISA project JSON file in a text editor (or a JSON viewer) and ensure that the src fields contain relative path (e.g. video1/f00001.jpg) to the frames (i.e. files[0]['src'] = 'video1/f0001.jpg'). Copy the LISA html application (i.e. lisa.html) file to /data/fish/allframes/ folder and open the application in a web browser from this folder. Now, you should see all the images and you do not even need to set the path. This method works because web browsers allow access to files stored in the local disk based on relative path (i.e. video1/f00001.jpg).