Page History

...

1. From the Camera position drop-down list, select:
   1. Wall—objects are detected only if their lower part gets into the area of interest specified in the detector settings.
   2. Ceiling—objects are detected even if their lower part doesn't get into the area of interest specified in the detector settings.
2. From the Device drop-down list, select the device on which the neural network operates: CPU, one of NVIDIA GPUs, or one of Intel GPUs. Auto (default value)—the device is selected automatically: NVIDIA GPU gets the highest priority, followed by Intel GPU, then CPU.
   

   Note
   title Attention!
   We recommend using the GPU. It can take several minutes to launch the algorithm on the 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 the Detector Pack subsystem of version 32.0, the support of Intel HDDL is removed, therefore, when you update from version 1.0, the Not supported option is selected automatically instead of this device variant, and detectors don't operate. To resume the detector operation, select the required device from the list.

3. From the Object type drop-down list, select the object type:
   • Human—the camera is directed at a person at an angle of 100-160°;
   • Human (top-down view)—the camera is directed at a person from above at a slight angle;
   • Vehicle—the camera is directed at a vehicle at an angle of 100-160°;
   • Person and vehicle (Nano)—person and vehicle recognition, small neural network size;
   • Person and vehicle (Medium)—person and vehicle recognition, medium neural network size;
   • Person and vehicle (Large)—person and vehicle recognition, large neural network size;
   • Human top-down view (Nano)—the camera is directed at a person from above at a slight angle, small neural network size;
   • Human top-down view (Medium)—the camera is directed at a person from above at a slight angle, medium neural network size;
   • Human top-down view (Large)—the camera is directed at a person from above at a slight angle, large neural network size;
     

     Info
     title Note
     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. The larger the neural network, the higher the accuracy of object recognition.

4. By default, the standard (default) neural network is initialized according to the object type selected in step 14 and the device type in step 13. The standard neural networks for different processor types are selected automatically; you must not do it manually. If you use a custom neural network, then click the button to the right of the Tracking model field, and in the standard Windows Explorer window that opens, specify its file
   

   Note
   title Attention!
   To train a neural network, contact the AxxonSoft technical support (see see Data collection requirements for neural network training). A neural network trained for a specific scene allows detecting objects of a certain type only (for example, a person, cyclist, motorcyclist, and so on).

5. Set the Scanning mode checkbox to detect small objects. If you enable this mode, the load on the system increases. So in step 10 we recommend specifying a small number of frames processed per second. By default, the checkbox is cleared. For more information on the scanning mode, see Configuring the Scanning mode.
6. Set the Model quantization checkbox to enable themodel quantization. By default, the checkbox is cleared. This parameter allows reducing the consumption of the GPU's computational power.
   

   Info
   title Note
   AxxonSoft conducted a study in which a neural network model was trained to identify the characteristics of the detected object. 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 percentage ranges within ±1.5%, and the difference in object identification ranges within ±2%. Model quantization is only applicable for NVIDIA GPUs. The first launch of the detector with the activated quantization feature can take longer than a standard launch. If you use the GPU caching, next time the detector with quantization runs without delays.

7. If necessary, specify the class of the detected object in the Target classes field. If you want to count and 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 view), 10—Vehicle.
   

   Info
   title Note
   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 counted and displayed (Object type, Neural network file). If you specify a class/classes missing from the neural network, tracks aren't counted and displayed.

8. In the Recognition threshold [0, 100] field, enter the neurocounter sensitivity—an integer value in the range from 0 to 100. The default value is 30.
   

   Info
   title Note
   The neurocounter sensitivity is determined experimentally. The lower the sensitivity, the higher the probability of false alarms. The higher the sensitivity, the lower the probability of false alarms, however, some useful tracks can be skipped. See Example of configuring neurocounter for solving typical tasks.

9. In the Number of frames for analysis and output field, specify the number of frames that must be processed to determine the number of objects in them.
10. In the Frames processed per second [0.016, 100] field, specify the number of frames processed per second by the neural network in the range from 0.016 to 100. For all other frames, interpolation is performed—finding intermediate values by the available discrete set of its known values. The greater the value of the parameter, the more accurate the detector operation, but the higher the load on the processor.

...