Configuration of the VI abandoned objects detection module includes general settings and configuration setup: determining the configuration parameters and detector characteristics.

General settings of the detector

1. Go to the settings panel of the VI abandoned objects detection object that is created on the basis of the Camera object on the Hardware tab of the System settings window.
   
2. In the Frames processed per second [1, 100] field, specify the number of frames in the range 1–100 that the detector processes per second. The default value is 12 (recommended). You can specify only a positive integer. If you enter a number outside the range 1–100, it is automatically changed to the nearest border value. If you leave the field blank, it automatically returns to the default value when you save the settings.
3. In the Scale frame to (pix.) field, specify in the range 480–1920 the width of the frame in pixels after scaling. The height of the frame is calculated automatically as the value specified in the setting, divided by 1.77777(7). The default value is 640, that is, the frame size after scaling is 640x360. The setting doesn't affect the resolution of the video stream and is used to reduce the load on the CPU, it is displayed in the logs and in the debug window.
4. Set the Use GPU checkbox to use NVIDIA GPU when working with the neural network.
5. Set the Show objects on image checkbox to outline the detection zone with a red border on the Video surveillance monitor when an object detection event is received.
6. From the Detection filtering drop-down list, select the filter that you want to use: Object filter, Object counting filter, Do not use filter (default). You can configure filters on the Postprocessing tab of the Detection settings window.
7. Click the Configuration setup button. The Detection settings window opens.

The Select area tab

1. In the Detection settings window, on the Select area tab, click the Stop video button to pause video playback.
   
   
2. By default, one detection zone is outlined with a red border. You can add more zones with the mouse, as well as change the borders of a zone by clicking its number in the list on the right. The number of a detection zone is highlighted in blue. 
   
   
3. To rename a detection zone, enter a new name in the field to the right of its number. The new name appears in the zone caption on the video image.
   

   To delete a detection zone, click the button to the right of the zone name. If there are any incorrect zones among the specified ones, after you save and reopen the settings, only the correctly specified zones remain. If you specify only incorrect zones, they are automatically replaced with a default zone after you save and reopen the settings. After you delete all zones, save and reopen the settings, a default zone is automatically added.

4. To save all changes, click the Apply configuration changes button.
5. To start video playback, click the Start video button.

The Parameters tab

1. Go to the Parameters tab. The parameters of the configuration that you use are displayed here. If you don't want to change the configuration, skip steps 2 and 3 and go to the next tab.
   
2. To use your configuration, click the Load configuration from file button.
3. In the standard Windows search window that opens, specify the path to the prepared configuration file in JSON format.

The Abandoned objects detection tab

1. Go to the Abandoned objects detection tab.
   
2. Clear the B/w mode checkbox to disable the black and white mode. By default, the checkbox is set.
3. In the Initial detection sensitivity [0, 1] field, specify the general sensitivity of the detector to detect abandoned objects in the range 0-1. The 0 value—no object is detected. The 1 value—maximum sensitivity. The default value is 0.70.
4. In the Sensitivity of primary detection in the foreground [0-1] field, specify the sensitivity of the detector to detect abandoned objects in the foreground in the range 0-1. The 0 value—no object is detected. The 1 value—maximum sensitivity. The default value is 0.70.
5. In the Object visibility [0, 1] field, specify in fractions the sensitivity of the detector to detect barely visible, merging with the background objects in the range 0-1. The higher the value of the parameter, the less noticeable an object can be detected. The recommended value (default) is 0.95.
6. In the Passenger traffic density [0, 1] field, specify the sensitivity of the object detection in conditions of intensive overlapping of the object by people passing in the camera FOV in the range 0-1. The default value is 0.99.
7. In the Minimum size of detected object field, specify in centimeters on one side the minimum allowable size of an object that you want to detect. You must specify the size of an object relative to the frame. Smaller objects don't trigger the detector. The default value is 20.
8. In the Maximum size of detected object field, specify in centimeters on one side the maximum allowable size of an object that you want to detect. You must specify the size of an object relative to the frame. Larger objects don't trigger the detector. The default value is 185.
9. In the Dead time (s) field, specify the time interval in seconds during which the occurrence of a repeated event is blocked. An event is considered repeated if its event intersection level is equal to or greater than the value specified in step 10. The default value is 0. 
10. In the Events intersection level field, in the range 0-1, specify the level representing the way the repeated events are filtered. An event is considered repeated if the ratio of the intersection area to the area of ​​the conjunction of two events is greater than or equal to the value specified for this parameter. The default value is 0.01. This parameter value is used in the Dead time parameter.
11. In the Detection time (s) field, specify the time interval in seconds after which an object is considered abandoned. The default value is 120.
12. Clear the No person checkbox so that objects near people are also detected. By default, the checkbox is set, which means that objects near people are ignored.
13. Clear the Filter objects checkbox to disable the neural network filter of objects. By default, the checkbox is set, which means that the neural network filter is enabled.
14. Clear the Background loading checkbox to disable the loading of the video background. By default, the checkbox is set. After restarting, the detector loads the previously saved background, not to retrain the network model. 
15. Clear the Background saving checkbox so that the background isn't saved. By default, the checkbox is set, which means that the background of the video is saved regularly with the periodicity specified in the Background saving period field. When you shut down the detector, the previously accumulated background is saved to a file for subsequent loading at startup.
    

    If you use the Background loading and Background saving parameters during initial configuration, you save time on automatic training at the first startup. Even if the checkboxes are clear, the background is saved and updated over the specified period, so to completely disable background saving, specify 0 in the Background saving period field.

16. In the Background saving period field, specify the time period in seconds for which the background of the video is saved. The default value is 0. 
17. In the Number of analytics streams field, specify the number of video streams that are used for the operation of the neural network analytics. The default value is 4.

The Additional settings tab

1. Go to the Additional settings tab.
   
   
2. Configure the false alarm filtering:
   1. From the Neural network filter mode for "phantom" triggers drop-down list, select which operation mode of the neural network filter mode for "phantom" triggers (false detections that appear when the detector operates on empty areas of the frame) you want to use:
      1. Precise (default) works slower, puts more load on the system, but the probability of the "phantom" triggers is significantly decreased;
      2. Fast works faster, puts less load on the system, but the probability of the "phantom" triggers increases.
   2. In the Power of neural network filter for "phantom" triggers field, specify the filtering level of the monotonous or monochromatic triggers, for example, spots on the floor connected with lighting change in the range 0-100. The value of the parameter is selected empirically, the recommended (default) value is 8 for the precise mode and 0.3 for the fast mode. If you change the mode, the optimal value is set automatically.
   3. In the Power of filter for "phantom" triggers field, specify the filtering level of shadows and flares depending on the scene contrast, lighting change, presence of shadows and flares. The value of the parameter is selected empirically, the default is 1.
3. Perform the advanced settings:
   1. In the Training period (s) field, specify the time in seconds during which the detector analyzes the scene. The longer the training process, the better the subsequent analysis. The recommended duration is from 15 to 30 seconds, 25 by default. The training quality depends on the "clarity" of the scene when the analytics is enabled: if the scene is noisy or the passenger traffic is too dense, we recommend increasing the period up to 5 minutes.
      There are three modes of configuring the training period:
      1. Automatically—automatic selection of the training period depending on other settings, equal to the -1 value;
      2. Manually—training period is set manually;
      3. Disabled—detector training is disabled. Equal to the 0 value.
   2. In the Post-training period (s) field, specify the time in seconds during which the detector "adapts" to the scene. If you disable the parameter, the false triggers can occur during the first minutes of the detector operation due to the algorithm peculiarity. It doesn't affect the quality of the subsequent analysis. There are three modes of configuring the post-training period:
      1. Automatically—automatic selection of the post-training period depending on other settings, equal to the -1 value;
      2. Manually—post-training period is set manually;
      3. Disabled—detector post-training is disabled. Equal to the 0 value.
   3. In the Frame sampling period (ms) field, specify the frame sampling rate for training in milliseconds. The default value is 67.
   4. In the Sampling rate [32, 256] field, specify the allowable margin of error for training results. The lower the value, the worse the training results. The recommended value is from 85 and greater. The default value is 128.
4. Click the OK button to save the settings of the detector.
5. To save the changes, click the Apply button on the settings panel of the VI abandoned objects detection object.

Configuration of the VI abandoned objects detection module is complete.