AI-CROWD-DEEP

The performance is not adequate, what else can I configure?



WARNING: Please read the disclaimer to make sure that all of the installation constraints have been fulfilled!

Go back


Installation Constraints

A camera usable for estimating the crowd with AI-Crowd DEEP must respect the following constraints:

  1. It must be installed in such a way that the framed people have a minimum height of 30 pixels at a resolution of 512x288 (a person height is at least 1/10 of the image height).
  2. If necessary, it must be installed with external illuminators that make it possible to distinguish the targets in all natural or artificial lighting conditions
  3. It is necessary to avoid drawing sensors under the camera, since the head and the shoulders may be not sufficient to recognize a person. The camera must be installed in a way that it provides a frontal/side view of the people. The use of cameras pointed towards the ground, giving a view from the top is not supported. The body of the people must be visible for at least 3/4 of its height.

Go back


Sensor configuration

Crowd estimation

Crowd Estimation

In the drop-down menu on the left, by selecting "Algorithm parameters" and then “Crowd” you can add, remove and edit sensors for crowd estimation within the scene framed by the camera.
Specifically, by clicking on “Add sensor” you can draw a virtual coil inside the image through "Redraw the sensor".

Redraw the sensor Using the pencil button next to the label to draw the polygon The sensor must be drawn on the ground: the people whose feet are inside the sensor will be counted

Sensor name The name of the area that will be shown by the event aggregator (e.g. A.I. Dash) Social Distance Social Distance Additional parameters:

  1. Sensor selection: identification of main sensor (only accepts whole numbers and it is not necessary to edit the default setting)
  2. Events Period (s): time interval (in seconds) between two consecutive events generated by the plugin.

Overcrowding detection

Overcrowding detection To view the result of the AI-Overcrowd module through the events mechanism, it’s necessary to check “Enable Overcrowd”. When enabled, the algorithm will evaluate all the Crowd sensors areas.

The main parameter to be configured is: Overcrowd threshold: If the number of people in the region of interest exceeds this threshold, the application raises an overcrowding alarm.

Further parameters that may be set are latency, confidence and inhibition.

Social distance

Social Distance Create a polygonal sensor on the ground for each area to be monitored: Click the “Add sensor” button to create a new sensor. The main parameters are:

Redraw the sensor Using the pencil button next to the label to draw the polygon The sensor must be drawn on the ground: the people whose feet are inside the sensor will be considered

Sensor name The name of the area that will be shown by the event aggregator (e.g. A.I. Dash)

Social distance threshold (m) Maximum allowed distance between

Social Distance Social Distance

Additional parameters:

  1. Latency (s): Time to wait in order to confirm the alarm condition before the event is triggered
  2. Confidence : The application generates an event if detects the alarm condition between two people for a certain percentage of the latency interval. This percentage is the confidence value.
  3. Inhibition (s): To avoid the constant generation of the state change, the system allows to determine an inhibition timespan, which is the minimum interval of time between two consecutive events generated by the same person.
  4. Sensor Id (Sensor selection): Unique id of the sensory area that will be sent along with events.

The sensor has three different operating modes that can be set through the Event generation parameter:

  1. One per person: In this mode an event will be generated for each person who violates the minimum distance.
  2. Global: When this mode is chosen the People threshold field is displayed on the screen and must be set for the correct use of this mode. In this mode an event will be generated when the number of people inside the sensor that violates the distance is greater than or equal to the value set for the People threshold field.
  3. Group based: This mode also makes use of the additional field People threshold. In this mode, an event will be generated when inside the sensor there is a group of people close to each other whose number is greater than or equal to the value set for the People threshold field.

Go back