|Training Note Video Analytics Human Behaviour|
This note explains the use of video analytics to detect a few scenarios of human behaviour that may need follow-up actions by another party. The scenarios covered here (as listed below) were classified as iQ-115 to reflect the intelligence level of the application.
-Slip and fall
The minimum hardware is a video surveillance camera connected to a PC based system that runs a Video Analytics software application. The VA application has to be set up in advance correctly. These are the essential and minimum set up:Area of interest which could be any polygon inside the video frame
Maximum size of target as a rectangle including some margins
Minimum size of target as a smaller rectangle
Slip and Fall
detects human slip and fall in public or private places such as retirement
homes or hospitals. When slip or fall is
detected, the system will send an alarm or alert message to another party for
follow up action. The message will
include a snapshot of the slip or fall so that the other party can decide what
to do next. To reduce alarming nuisance,
the application will ignore slip or fall that take place very slowly such as
people bending to tie shoe lace or stretching when tired. This requires a set up in advance of a short time
period (several seconds) for qualifying motions.
Loitering can be harmful to public or private safety or security because criminals loiter a while before launching an attack. Loitering will certainly be picked up by a surveillance camera on motion detection basis among all other situations such as environmental illumination intensity change without discrimination. Detection of loitering does require special treatment in terms of a longer time period (such as 15 seconds) for qualifying motions. This may not work well in a crowded scene as obstructions by a different object will reset the clock to count again.
Run and Speed
Someone is rushing (moving faster than normal). It can be a robbery. It can be the start of a fight. It can be many other things too. Irrespectively we can look at what has happened and decide if to respond or not. To detect running, the area of monitoring must be several times of the distance of a long step so that the application is able to compute the speed of movement.