The sighting duration is in proportional relation to the time period in which the system is performing detections of the identity on the given camera. Therefore, the longer the identity stays in front of the camera the longer will the sighting last and the bigger number of detections will be. There is one exception to this rule. There are certain "grace periods", a functionality implemented in two ways.
1. Sightings of the same identity are grouped as one sighting if the previous sighting ended less than 60 seconds before the current sighting started. The sixty-second gap is a configurable threshold. In this case, grace periods are designed to decrease the number of needless sightings.
2. Also, we have a functionality to stop sighting concatenation if the last sighting started more than 10 minutes before the current sighting started. 10 minutes gap is also a configurable parameter. This is to prevent long-lasting sightings that could possibly be generated if the person is always visible in front of the camera.
Each initial detection triggers a new sighting. The sighting is the sequence of detections of the same identity from sequential frames. The sighting collects all the detections inside as long as the given identity is in the corresponding physical area with corresponding dimensions.
As soon as the system detects the face for which there is no correspondent already captured detection in earlier frames, a new sighting is initialized.
Sighting duration is directly related to the stream quality and camera-friendly environment that is being recorded. If those two conditions are not met, AI services will have troubles with execution and quality rate.
For example, if the frame rate is low, one subject can spend a longer time in front of the camera and be given three sightings instead of one. There is a possibility that the subject won't even be given identity by classifying sighting as a sighting w/o identity. In other words, the higher the quality of the stream higher the quality of sightings.