Capable of tracking objects that are stationary for short periods of time.

For example, use in areas where you want to track people standing still for 30 to 60 seconds.

Excellent all-purpose Indoor People Tracking background model.

Uses color, brightness, and texture information. This background model can also use multi-scale processing to fuse information from different scales.

You can customize to different scenarios by adjusting the sensitivity to intensity, color, or texture information in a given scene.

Controls how aggressively shadows (decreases in lighting) are ignored when tracking moving objects. Disabled by default.

When enabled, increase the slider (to the right) to allow the engine to correctly ignore more of the shadow areas.

Note: Ignoring more shadow areas increases the potential of a person wearing dark clothing on a light background to be incorrectly classified as a shadow.

Controls how aggressively increases in lighting are ignored. Disabled by default.

When enabled, increases in lighting occur due to a light source intensifying, such as a car headlight beam shining on an area, or an overcast day becoming a sunny day.

Increase the slider (to the right) to allow the engine to correctly ignore any increases in illumination.

Note: Ignorning lighting increases also increases the potential of a person wearing light clothing on a dark background to be incorrectly classified as an illumination increase.

Tracks an object moving very slowly across the video, or an object coming directly toward or moving directly away from the camera (which will appear to be moving slowly).

Disabled by default. When disabled, slow moving objects tend to become part of the background and will not be tracked.

When enabled, slow moving objects will be tracked.

Important: This option might increase some false alarms for situations where an object stands still in the video for a long time.

Timing

The Dual-Layer Dynamic and Multi-Layer Hybrid background models interprets a scene using two representations: short-term and long-term. Short-term representation are objects in the scene that have recently moved or just arrived in the scene. These are the objects that will be tracked.

Long-term representation are background objects that are not going to be tracked. Occasionally, some objects will move from short-term representation to long-term representation. Also at some point the short-term representations and long-term representations must be deleted because it has changed over time.

The time to delete a long-term representation that is not currently visible.

Example scenario using a car parking lot:

A car left the parking spot. How long do you want the analytic to remember that it was there? If the same car comes back and parks in the same spot, and the old background is still remembered, then it will just be background instantly.

The time in which stationary objects are reported as tracked objects, or the expected maximum dwell time for stationary objects.

Considerations:

Sometimes the background is changing between several states, for example, like a neon sign switching on an off. If the period of the oscillation is considered (seconds), the analytic can potentially learn both lighting states as background. On the other hand, if you have a scene where people in dark clothes are always passing by in the same spot and the time to clear is set for too many seconds, the dark clothes at this location might be seen as an intermittent feature of the background.

The time to delete the representation of objects that pass through the scene.

Many objects will move through the scene over time. The analytic learns these objects as part of the short-term representation. However, once they move (away from a spot or away from the scene), these past representations can be cleared from memory.

The background model tracks consistent motion, automatically detects abnormal behavior, or tracks motion only in a specified direction as follows:

• Coherent motion – Tracks consistent motion; therefore, greatly reduces false alarms due to inconsistent motion such as tree branches and other objects in the background that can move randomly due to the wind. Set by default.
• Abnormal behavior – Background model learns the normal patterns and directions of motion at each pixel; therefore, any abnormal direction of motion will be detected.
• Wrong direction – Tracks motion only in a specified direction; therefore, motion in any other direction will be ignored.

Timing - see the Dual Layer Dynamic Background Model descriptions for details on:

Time to clear long-term representation (seconds)
Time to clear short-term representation (seconds)
Time to move long-term representation (seconds)

Controls how sensitive the background model is to perceived changes from the expected background values. Evaluates up to four (4) features (and at least one feature should be selected):

• Brightness (from black to white)
• Thermal
• Color (red, blue, gray, etc.)
• Texture (local patterns of brightness)

If sensitivity is low, changes are attrbuted to natural variations in the appearance of the background. If sensitivity is high, changes are attributed to a foreground object.

Looks for changes in gray-level (from black to white).

Enabled by default, but the appropriate sensitivity depends on how much the lighting in the scene varies over time.

The engine looks for change in hue and saturation. Enabled by default.

Color is usually not as strongly affected by lighting, but not all objects can be distinguished from the background by color alone.

This option requires more CPU time than Brightness.

The engine looks for changes in the local brightness pattern, especially new edges. Disabled by default.

Texture is generally less affected by lighting, but flatter objects may not have enough texture to be distinguished from the background.

This option requires more CPU time than Color.

Select this option to monitor for changes at multiple spatial resolutions. Disabled by default.

This option can improve accuracy for difficult scenes (especially in combination with Texture features), but it also increases CPU load.

Adaptation

Time to adapt representations to show changes (seconds)

Controls how quickly the background model can adapt to slow changes in the scene (such as the sun going down).

Use a large value if you need to detect slow changes in the scene.

Use a small value if the scene contains many gradual but relatively quick changes in lighting.

Defines how much an object can move and still be considered to be dwelling. Horizontal and Vertical are calculated as a percent of object size that the object can move.

Examples:

In an Uncalibrated Camera Position (via the Overview subtab), Horizontal 50 means an object can move up to 50% of its max height & width horizontally before it is considered to have moved. This is similar for Vertical.

For calibrated Camera Position (Angled, Overhead), Horizontal sets the percentage of height that the object can move on the ground plane. Vertical is ignored.

Defines how much analysis the algorithm is allowed, measured in frames.

Indoor only

Defines how much analysis the algorithm is allowed, measured in objects (if there are few objects).

Indoor only

Defines how willing the system is to think about moves that do not immediately improve the scene model’s correspondence with the available evidence.

Indoor only

Defines how far the object needs to travel before the user sees it appear on the screen within the camera view.

For example, a stationary tree may sway in a breeze, but its movement would not register the tree as an object to track so long as it is moving a distance smaller than the New Object Min Travel Distance value.

Defines how long to keep a previously tracked object hidden in the scene.

Keeping the object hidden can allow for re-tracking that same object when the object becomes occluded for a short period of time.