VE180 - Motion Tracking

Environment

•       Away from moving objects such as elevator doors, or moved objects such as carts, carriages and large luggage.

•       Avoid camera placement near escalators or stairs. Ensure that people are tracked on flat and level surfaces for accurate detection.

•       Ensure that people are fully visible to the camera and not partially hidden by shelves or low walls.

Lighting

•       At least 75 LUX or better.

•       Ensure there is consistent light throughout the space and throughout the period of time that people are tracked.

•       Ensure that there are no shiny objects, like a shiny floor in the area people are tracked. A shiny floor may reflect light/shadows from surrounding area and generate false counts.

•       In the indoor tracking area, if possible, avoid sunlight from outside. Sunlight occasionally obscured by moving clouds will cast shadows that cross a counting line.

•       Depending on the angle of an indoor light, the shadows of people walking by (but not through) a counting line can trigger false counts.

Counting Lines

•       An overhead view is usually more accurate for counting people across a line, but it covers less ground than a 45-degree view if the ceiling is low.

•       Try to have at least 1 meter of floor space on either side of the counting line.

•       In angled views, try to have at least 2 meters of space on either side of the counting line.

•       Setting person size: A person crossing the line should not take up more than about 1/3 of the view in width and no more than 1/2 of the view in height.

Camera Placement

•       Note that accuracy decreases for nearly horizontal views (for example, less than 20 degrees declination).

•       If the camera is placed such that people appear smaller than 20 pixels on the screen, counting accuracy will decrease and you will need to use advanced option settings.

For Outdoor Multi-Class Object Tracking

1. From the Task drop-down list, select Outdoor Multi-Class Object Tracking.

2. From the Camera Position drop-down list, select Angled, Overhead, Uncalibrated or Auto Calibration (BETA).

Uncalibrated - Use only if objects are well separated and accurate counts are not required.

3      From the Environment Type drop-down list, select one of the following.

Outdoor: Generic

Outdoor: Water

Outdoor: Thermal camera

Custom - Automatically displayed if settings have been changed from default values in the Environment tab or Advanced settings.

4        From the Performance drop-down list, select the level of CPU processing: Custom, Low, Intermediate, High. By default, Intermediate is selected.

When you select the Outdoor Multi-Class Object Tracking task, the Outdoor Background Model is selected by default on the Environment subtab. See the Environment subtab section for further details on available configuration options.

For Face Tracking

From the Task drop-down list, select Face Tracking.

No additional options are available on the Overview subtab.

See the sections detailing the Processing Mask subtab, Counting Line subtab, and Advanced subtab configuration options.

Grid

Grid lines are 1 meter apart.

Automatically displayed over scene.

Reference person icon

An orange colored form representing person size and position in an image.

Four populate the image automatically.

Move the reference people icons to various positions in the image, foreground and background as they would naturally occur.

To add more reference person icons, click your mouse on the image.

Delete Person button

Removes a reference person icon.

Click the reference person icon and then click the Delete Person button.

Hide/Show World button

Hides (or displays) the grid lines and reference person icons so that you can view the actual scene in the image.

Interactive section

Allows you to adjust the grid and reference person size to match the underlying scene.

Using your mouse, you drag the camera icon up and down and left and right.

+/- magnification icons

Allow you to zoom in and out in the interactive section

Field of view (FOV)

Horizontal field of view.

If you increase the value, you must drag the camera icon closer (down) to get the correct person size.

If you decrease the value, you must drag the camera icon further away for the same rough size.

If you obtain the exact camera position by measuring the actual height and distance, then you must adjust Field of view to make the reference people icons the right size in the scene.

Mouse wheel perspective control

Allows you to control the amount of perspective in the scene without changing the overall scale.  (Allows you to make a person icon a certain size while changing the Field of view so that the tilt of the other people icons and their relative size is correct.)

Position the mouse over the little diagram. Rotate the mouse wheel to increase or decrease the Field of view in 2.5 degrees increments, creating an opposite proportional change in the ground distance parameter.

Example: Ground distance is 10.0 meters and the Field of view is 60 degrees. One mouse wheel click changes Field of view to 62.5 degrees and ground distance to (60/62.5)*10.0 = 9.6 m. A mouse wheel click in the other direction changes Field of view to 57.5 degrees and ground distance to (60/57.5)*10.0 = 10.43 m.

Pan

Movement of the camera from left to right or right to left around the imaginary vertical axis that runs through the camera.

When the camera is pointed at the ground, it basically tilts the ground plane. When the camera is closer to horizontal, it looks more like a rotation of the ground plane.

Here Pan tilts the grid against the ground plane, which is incorrect based on the actual scene.

Most useful in an near-overhead view when the camera is pointing off to the side (that is, the deflection from vertical is not completely on the camera view’s up-down axis)

Twist

Rotation of the camera around the axis that extends directly out of the camera through the center of the view.

Useful in an angled view when the camera is twisted a bit so that a vertical line in the world does not correspond to up-down in the image. Also useful if the ground in the image appears to tilt a bit from left to right

Height and Distance fields

Values correspond to the movements of the camera icon in the interactive section.

Warning message

Symphony displays a warning message if the camera angle does not agree with Overhead selection

Background learning duration (seconds)   

Determines the approximate length of time it will take a foreground object that becomes stationary to merge into the background.

Important: If an object has periodic movement with a period greater than the specified time, then that object will always be considered a foreground object (never merged into the background).

The background is updated all the time, not only when the analytic engine starts.

Background Sensitivity   

Sets the minimum background sensitivity needed for an object to be detected as a foreground object.

Important: If set to a low value, then objects with similar color as the background will NOT be detected as valid objects to track.

Learn normal behavior

Learns the percentage of time each pixel is in motion and can subsequently ignore. Useful, for example, when tracking boats on water. Disabled by default.

Method

•       % of screen change: Monitors the percentage of the scene that is considered foreground. If the amount of foreground is too high, model relearning is triggered.

•       Brightness change: Monitors the scene for abrupt, overall changes in brightness (perhaps caused by a change in lighting or weather event).

•       Spectrum change: Monitors the scene for abrupt, overall changes in brightness, but less sensitive to local changes (such as a person in a dark coat walking by).

•       Combined change: A combination of % of screen change and Spectrum change.

Sensitivity

Value range between 0 and 100.

A high sensitivity means the relearning is easily triggered. For example, 80 sensitivity will trigger when just 20% of the screen is foreground, while a low sensitivity triggers relearning only in an extreme situation.

Time to wait before resuming tracking (seconds)

Allows you to set how long to wait before resuming to track objects.

Ideally this value should be zero, but if you are aware of a lighting issue where tracking immediately can cause false alarms, then set this to the number of seconds needed for the light to stabilize before tracking starts.

Shadow/Illumination Removal

Disabled by default.

Strength

A high sensitivity level setting is helpful in cases where you have a very strong dark shadow. Disabled by default.

A low sensitivity level setting is help in cases where you have a light shadow.

Example:

0.1 = a light shadow

1.0 = a very strong dark shadow

Snow Detection

Enable this option to configure the background model to detect and track in areas covered by snow. Disabled by default.

Detect Small Objects

Enable this option to configure the background model to detect and track the smaller-scale objects. For example, an office tower camera that covers a large pedestrian traffic area would contain uniformly small objects in the camera view.

Simple

Enabled by default. Uses less CPU than Complex. Works for most situations.

Complex

In cases where there are strong shadows, Complex removes more shadows, but uses more CPU.

Strength

Defines how aggressively to remove shadows.

•       Increasing the strength value can result in an increase in potential objects being missed.

•       Decreasing the strength value can result in shadows being tracked as objects.

Track Slow Moving Objects

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.

Object Speed

Categories the speed of the tracked object.

Object Contrast

Controls how sensitive the background model is to perceived similarities between the background and objects. If an object in the camera view appears similar to the background (such as a dark background and a person in dark clothing), increasing the object contrast threshold increases the model’s ability to detect the object against the dark background.

Method

•       % of screen change: Monitors the percentage of the scene that is considered foreground. If the amount of foreground is too high, model relearning is triggered.

•       Brightness change: Monitors the scene for abrupt, overall changes in brightness (perhaps caused by a change in lighting or weather event).

•       Spectrum change: Monitors the scene for abrupt, overall changes in brightness, but less sensitive to local changes (such as a person in a dark coat walking by).

•       Combined change: A combination of % of screen change and Spectrum change.

Sensitivity

Value range between 0 and 100.

A high sensitivity means the relearning is easily triggered. For example, 80 sensitivity will trigger when just 20% of the screen is foreground, while a low sensitivity triggers relearning only in an extreme situation.

Time to wait before resuming tracking (seconds)

Allows you to set how long to wait before resuming to track objects.

Ideally this value should be zero, but if you are aware of a lighting issue where tracking immediately can cause false alarms, then set this to the number of seconds needed for the light to stabilize before tracking starts.

Analysis FPS

Indicates how many frames will be analyzed by the system.

Typically the default value should not be altered. However, in many cases the analysis frame rate can be reduced to conserve CPU. Doing this may negatively alter the algorithm’s performance.

Capture Resolution

Displays the resolution at which the video is being recorded. This can be changed in the Device setup.

Analysis Resolution

Setting the Analysis Resolution the same as the Capture Resolution ensures that objects will be detected as far away as possible. However, in order to conserve CPU resources, it is possible to down sample the video sent for analysis. Downsampling will reduce CPU requirements but may also reduce the detection range of the video analytic.