Access to City Cameras

Post date: Dec 12, 2015 6:15:39 PM

This term we have been testing the possibility of using cameras to count people automatically. We have come to the realization that only 3 out of 9 cameras are accurate within the 15% margin of error. This unfortunate turn out could be due to a number of different problems like umbrellas, tourists moving across the line multiple times to take pictures, babies in carriages, but most importantly poor wifi connection and camera angle.

This past week, however, the team was given access to three city cameras. Access to these cameras gave the team the ability to isolate the poor wifi connection as the city cameras are hard wired. We were able to test solely based on the camera angle to come to the conclusion that a birds-eye, or overhead view is the most accurate view. This means point tilt zoom cameras, like the ones the city uses, are not viable for counting pedestrians.

Getting access to the city cameras was a step in the right direction. Isolating a major problem showed us just how viable of a solution this concept is. Overall we were very glad to find out that it is possible to use pre-existing cameras for the algorithm to count people.