Configuring the Neural tracker

On the page:

Configuring the detector

To configure the Neural tracker, do the following:

Go to the Detectors tab.
Below the required camera, click Create… → Category: Trackers → Neural tracker.

By default, the detector is enabled and set to detect moving people.

If necessary, you can change the detector parameters. The list of parameters is given in the table:

Parameter	Value	Description
Object features
Record object trajectories	Yes	By default, metadata are recorded into the database. To disable metadata recording, select the No value Attention! To obtain metadata, video is decompressed and analyzed, which places a heavy load on the server and limits the number of cameras used on it.
Record object trajectories	No
Video stream	Main stream	If the camera supports multistreaming, select the stream for which detection is needed
Other
Enable	Yes	By default, the detector is enabled. To disable, select the No value
Enable	No
Name	Neural tracker	Enter the detector name or leave the default name
Decoder mode	Auto	Select a processing resource for decoding video streams. When you select a GPU, a stand-alone graphics card takes priority (when decoding with Nvidia NVDEC chips). If there is no appropriate GPU, the decoding will use the Intel Quick Sync Video technology. Otherwise, CPU resources are used for decoding
	CPU
	GPU
	HuaweiNPU
Number of frames processed per second	6	Specify the number of frames for the neural network to process per second. The higher the value, the more accurate the tracking, but the load on the CPU is also higher. The value must be in the range [0.016, 100] Attention! We recommend the value of at least 6 FPS. For fast-moving objects (running individuals, vehicles), you must set the frame rate at 12 FPS or above.
Type	Neural tracker	Name of the detector type (non-editable field)
Advanced settings
Camera position	Wall	To sort out false events from the detector when using a fisheye camera, select the correct device location. For other devices, this parameter is irrelevant
Camera position	Ceiling
Hide moving objects	Yes	By default, the parameter is disabled. If you don't need to detect moving objects, select the Yes value. An object is considered static if it doesn't change its position more than 10% of its width or height during its track lifetime Attention! If a static object starts moving, the detector creates a track, and the object is no longer considered static.
Hide moving objects	No
Hide static objects	Yes	Starting with Detector Pack 3.14, the parameter is disabled by default. If you need to hide static objects, select the Yes value. This parameter lowers the number of false events from the detector when detecting moving objects. An object is considered static if it hasn't moved more than 10% of its width or height during the whole time of its track existence Attention! If a static object starts moving, the detector creates a track, and the object is no longer considered static. If you disable this parameter, the load on the CPU reduces. Starting with Detector Pack 3.15, the feature of accumulation of a background mask of static objects has been moved to the ENABLE_STATIC_OBJECTS_MASK system variable (see System variables for the Neural tracker).
Hide static objects	No
Minimum number of detection triggers	6	Specify the Minimum number of detection triggers for the Neural tracker to display the object's track. The higher the value, the longer the time interval between the detection of an object and the display of its track on the screen. Low values of this parameter can lead to false events from the detector. The value must be in the range [2, 100]
Model quantization	Yes	By default, the parameter is disabled. The parameter is applicable only to standard neural networks for Nvidia GPUs. It allows you to reduce the consumption of computation power. The neural network is selected automatically, depending on the value selected in the Detection neural network parameter. To quantize the model, select the Yes value Attention! AxxonSoft conducted a study in which a neural network model was trained to identify the characteristics of the detected object with quantization. The following results of the study were obtained: model quantization can lead to both an increase in the percentage of recognition and a decrease. This is due to the generalization of the mathematical model. The difference in detection ranges within ±1.5%, and the difference in object identification ranges within ±2%. The first launch of a detector with the Model quantization parameter enabled can take longer than a standard launch. If GPU caching is used, next time the detector with quantization runs without delay.
Model quantization	No
Neural network file		If you use a custom neural network, select the corresponding file. Attention! To train your neural network, contact AxxonSoft (see Data collection requirements for neural network training). A trained neural network for a particular scene allows you to detect only objects of a certain type (for example, a person, a cyclist, a motorcyclist, and so on). If you don't specify the neural network file, the default file is used that is selected automatically, depending on the selected value in the Detection neural network parameter and the selected processor for the neural network operation in the Decoder mode parameter. If you use a custom neural network, enter a path to the file. The selected detection neural network is ignored when you use a custom neural network. You cannot specify the network file in Windows OS. You must place the neural network file locally, that is, on the same server where you install Axxon One. For correct neural network operation on Linux OS, place the corresponding file locally in the /opt/AxxonSoft/DetectorPack/NeuroSDK directory or in the network folder with the corresponding access rights.
Scanning mode	Yes	By default, the parameter is disabled. To enable the scanning mode, select the Yes value (see Configuring the scanning mode)
Scanning mode	No
Scanning window height	0	The height and width of the scanning window are determined according to the actual size of the frame and the required number of windows. For example, the real frame size is 1920×1080 pixels. To divide the frame into four equal windows, set the width of the scanning window to 960 pixels and the height to 540 pixels Attention! If the set height of the scanning window exceeds the height of the initial video stream, the video stream height is applied automatically. The same rule is applied to the width. Example 1: the size of both windows exceeds the video stream. Script: the video stream resolution is 1920x1080, the set window size is 2500x2000 Result: the system automatically applies the 1920x1080 window size, as both set values (height and width) are greater than the corresponding size of the video stream. Example 2: the size of only one window exceeds the video stream. Script: the video stream resolution is 1920x1080, the set window size is 2500x900 Result: the system automatically corrects only the exceeding parameter. The 1920x900 window is applied where the width is taken from the video stream while the set height (900px) is lower than the stream height and remains unchanged.
Scanning window step height	0	The scanning step determines the relative offset of the windows. If the step is equal to the height and width of the scanning window, respectively, the segments are lined up one after another. Reducing the height or width of the scanning step increases the number of windows due to their overlapping each other with an offset. This increases the detection accuracy but can also increase the load on the CPU Attention! The height and width of the scanning step mustn't be greater than the height and width of the scanning window, since the detector doesn't operate with such settings.
Scanning window width	0	The height and width of the scanning window are determined according to the actual size of the frame and the required number of windows. For example, the real frame size is 1920×1080 pixels. To divide the frame into four equal windows, set the width of the scanning window to 960 pixels and the height to 540 pixels Attention! If the set width of the scanning window exceeds the width of the initial video stream, the video stream width is applied automatically. The same rule is applied to the height. Example 1: the size of both windows exceeds the video stream. Script: the video stream resolution is 1920x1080, the set window size is 2500x2000 Result: the system automatically applies the 1920x1080 window size, as both set values (height and width) are greater than the corresponding size of the video stream. Example 2: the size of only one window exceeds the video stream. Script: the video stream resolution is 1600x1080, the set window size is 1600x2000 Result: the system automatically corrects only the exceeding parameter. The 1600x1080 window is applied where the height remains unchanged while the set width (2000px) is greater than the stream width and is taken from the video stream.
Scanning window step width	0	The scanning step determines the relative offset of the windows. If the step is equal to the height and width of the scanning window, respectively, the segments are lined up one after another. Reducing the height or width of the scanning step increases the number of windows due to their overlapping each other with an offset. This increases the detection accuracy but can also increase the load on the CPU Attention! The height and width of the scanning step mustn't be greater than the height and width of the scanning window, since the detector doesn't operate with such settings.
Selected object classes		If necessary, specify the class of the detected object. If you want to display tracks of several classes, specify them separated by a comma with a space. For example, 1, 10 The numerical values of classes for the embedded neural networks: 1—Human/Human (top-down view), 10—Vehicle If you leave the field blank, the tracks of all available classes from the neural network are displayed (Detection neural network, Neural network file) If you specify a class/classes from the neural network, the tracks of the specified class/classes are displayed (Detection neural network, Neural network file) If you specify a class/classes from the neural network and a class/classes missing from the neural network, the tracks of a class/classes from the neural network are displayed (Detection neural network, Neural network file) If you specify a class/classes missing from the neural network, the tracks of all available classes from the neural network are displayed (Detection neural network, Neural network file) Note Starting with Detector Pack 3.10.2, if you specify a class/classes missing from the neural network, the tracks aren't displayed (Detection neural network, Neural network file).
Sensitivity of excluding static objects (starting with Detector Pack 3.14)	25	Specify the level of sensitivity of excluding static objects. The higher the value, the less sensitive to motion the algorithm becomes. The value must be in the range [0, 100]
Similitude search	Yes	By default, the parameter is disabled. To enable the search for similar persons, select the Yes value. If you enable the parameter, it increases the load on the CPU Attention! The Similitude search works only on tracks of people.
Similitude search	No
Time of processing similitude track (sec)	0	Specify the time in seconds for the algorithm to process the track to search for similar persons. The value must be in the range [0, 3600]
Time period of excluding static objects	0	Specify the time in seconds after which the track of the static object is hidden. If the value of the parameter is 0, the track of the static object isn't hidden. The value must be in the range [0, 86 400]
Track lifespan (starting with Detector Pack 3.14)	Yes	By default, the parameter is disabled. If you want to display the track lifespan for an object in seconds, select the Yes value
Track lifespan (starting with Detector Pack 3.14)	No
Track retention time (sec)	0.7	Specify the time in seconds after which the object track is considered lost. This helps if objects in the scene temporarily overlap each other. For example, when a larger vehicle completely blocks the smaller one from view. The value must be in the range [0.3, 1000]
Basic settings
Detection threshold	30	Specify the Detection threshold for objects in percent. If the recognition probability falls below the specified value, the data will be ignored. The higher the value, the higher the detection quality, but some events from the detector may not be considered. The value must be in the range [0.05, 100]
Neural tracker mode	CPU	Select the processor for the neural network operation (see Hardware requirements for neural analytics operation, Selecting Nvidia GPU when configuring detectors) Attention! We recommend using the GPU. It can take several minutes to launch the algorithm on an Nvidia GPU after you apply the settings. You can use caching to speed up future launches (see Optimizing the operation of neural analytics on GPU in Windows OS). If the neural tracker is running on the GPU, object tracks can lag behind the objects in the Surveillance window. If this happens, set the camera buffer size to 1000 milliseconds (see Camera). Starting with Detector Pack 3.11, Intel HDDL and Intel NCS aren’t supported. Starting with Detector Pack 3.14, Intel Multi-GPU and Intel GPU 0-3 are supported.
	Nvidia GPU 0
	Nvidia GPU 1
	Nvidia GPU 2
	Nvidia GPU 3
	Intel NCS (not supported)
	Intel HDDL (not supported)
	Intel GPU
	Intel Multi-GPU
	Intel GPU 0
	Intel GPU 1
	Intel GPU 2
	Intel GPU 3
	Huawei NPU
Detection neural network	Person	Select the detection neural network from the list. By default, the Person detection neural network is selected. Neural networks are named taking into account the objects they detect. The names can include the size of the neural network (Nano, Medium, Large), which indicates the amount of consumed resources (see Video stream and scene requirements for the Neural tracker and its sub-detectors). The larger the neural network, the higher the accuracy of object recognition
	Person (top-down view)
	Person (top-down view Nano)
	Person (top-down view Medium)
	Person (top-down view Large)
	Vehicle
	Person and vehicle (Nano)
	Person and vehicle (Medium)
	Person and vehicle (Large)
Neural network filter
Neural filter	Yes	By default, the parameter is disabled. To sort out parts of tracks, select the Yes value. For example: The Neural tracker detects all freight trucks, and the Neural filter sorts out only the tracks that contain trucks with cargo doors open
Neural filter	No
Neural filter file		Select a neural network file. You must place the neural network file locally, that is, on the same server where you install Axxon One. You cannot specify the network file in Windows OS Attention! Starting with Detector Pack 3.12, the neural network file of the neural filter must match the processor type specified in the Neural tracker mode parameter. If you use a standard neural network (training wasn't performed in operating conditions), we guarantee an overall accuracy of 80-95% and a percentage of false positives of 5-20%. The standard neural networks are located in the C:\Program Files\Common Files\AxxonSoft\DetectorPack\NeuroSDK directory.

By default, the entire frame is a detection area. If necessary, in the preview window, you can reduce the detection area (see Configuring a detection area) and/or specify one or more ignore areas (see Configuring the ignore area).

Note

For convenience of configuration, you can "freeze" the frame. Click the button. To cancel the action, click this button again.
The detection area is displayed by default. To hide it, click the button. To cancel the action, click this button again.

To save the parameters of the detector, click the Apply button. To cancel the changes, click the Cancel button.

Configuring the Neural tracker is complete. If necessary, you can create and configure the necessary sub-detectors on the basis of the neural tracker (see Standard sub-detectors).

Attention!

To get an event from the Motion in area sub-detector on the basis of the Neural tracker, an object must be displaced by at least 25% of its width or height in the frame.

System variables for the Neural tracker

Creating system variable

Variable	Starting with	Purpose	Value	Description
ENABLE_CALC_HSV	Detector Pack 3.14	Detect the color of an object	0	Disable color detection. When you select this value, the load on the CPU reduces, including when the detector operates in the GPU-Nvidia GPU 0, 1, 2, or 3 modes. By default, when you select GPU-Nvidia GPU 0, 1, 2, or 3 in the Decoder mode and Neural tracker mode parameters, the ENABLE_CALC_HSV system variable is set to 0
ENABLE_CALC_HSV	Detector Pack 3.14	Detect the color of an object	1	Enable color detection. The system collects data about object color. This data is required for further color-based archive searches (see Search in archive). When you select this value, the load on the server increases and limits the number of cameras used. By default, when you select CPU-CPU, CPU-Nvidia GPU 0, 1, 2, or 3, GPU-CPU in the Decoder mode and Neural tracker mode parameters, the ENABLE_CALC_HSV system variable is set to 1
ENABLE_STATIC_OBJECTS_MASK	Detector Pack 3.15	Detection of accumulation of a background mask of static objects	0	Disable accumulation (default). When you select this value, the load on the CPU reduces, even when you select the GPU value in the Decoder mode parameter
ENABLE_STATIC_OBJECTS_MASK	Detector Pack 3.15		1	Enable accumulation. This value improves the quality of hiding static objects (the Hide static objects parameter). When you select this value, the load on the server increases

Example of configuring Neural tracker for solving typical tasks

Parameter	Task: detection of moving people	Task: detection of moving vehicles
Other
Number of frames processed per second	6	12
Neural network filter
Neural filter	No	No
Basic settings
Detection threshold	30	30
Advanced settings
Minimum number of detection triggers	6	6
Camera position	Wall	Wall
Hide static objects	Yes	Yes
Neural network file	Path to the .ann neural network file. You can also select the value in the Detection neural network* parameter. In this case, this field must be left blank	Path to the .ann neural network file. You can also select the value in the Detection neural network* parameter. In this case, this field must be left blank

Page tree

Configuring the Neural tracker

Configuring the detector

System variables for the Neural tracker

Example of configuring Neural tracker for solving typical tasks