page 20
TrafficDrone: How to start
Camera framing
Camera positioning and framing are defining parameters for video stream quality which
is the main driving factor of AI traffic analytics accuracy. Therefore, we do recommend to
pay appropriate attention to fine-tuning the acquisition geometry. Here are a few points
to consider:
●
Camera image center should be ideally identical with Region of Interest center
●
There should be sufficient padding between the image borders and the Region of
Interest (ROI)
●
Incidence angle between the camera lens axis and the ground plane normal should
be ideally 0-45 degrees in order to avoid reflections and dynamic occlusions. Top-
down view (incidence angle 0°) allows for maximum accuracy.
●
The objects in ROI should be fully visible; at least 70 % of the object area should
be visible to achieve high accuracy (be aware of dynamic/static occlusions)
●
Do not capture the sky - the sun can have negative impact on video analysis (tilt
the camera down if necessary)
●
If possible, orient the camera perpendicular to the traffic flow or aligned with the
flow, in the direction of travel - this helps avoiding reflections from vehicle lights
in night scenes.
Region of Interest
