The output is the final step of the processing pipeline, where the recognition results are sent to a storage or trigger certain actions. The pipeline allows several outputs working together. These are the platform's standard output types.


The System_DB output sends detected objects together with all their properties to the local DB. This is a default output. Saved data will be available in the dashboards.

system db output

There is a number of parameters that controls, which events to save:

  • min_confidence: DB will accept only the objects with the recognition confidence above this value.
  • min_track_length: This is an important parameter that allows you to exclude sketchy objects that were detected and then disappeared in next few frames. For example, if min_track_length equals 10 the object has to be on the scene long enough to be detected 10 times at least before system allows it to DB.
  • without_images: Each event can be saved with or without a reference image. It is important to save reference images, when the system is calibrated. After that saving the images can be switched off to save network traffic and hard drive space.

DB records events, related to detected objects. The switchers save_best_shot, save_last_shot and save_newly_detected control, which of standard events to save. Special events, such as line_cross, always recorded.


This output sends notification to the web clients if certain event occurs. User receives corresponding messages as pop-up notifications in the browser. By default all users are subscribed to the output_notifications. Therefore as soon as a Head meets a certain event and sends notification, the Manager distributes it across all users. Each user can unsubscribe from this type of notifications in the settings menu.

notification output

Use these controls to define the parameters of the notifications.

  • custom_message: This custom text will lead the notification.
  • event_filters: You can use up to three filters in addition to the object_name. Format is key: value. For example, direction: forward or age: >30. Values support < > = symbols.
  • min_confidence: The object recognition confidence should be above this value.
  • min_message_interval: This parameter protects users from an avalanche of messages. It forces the system to wait this amount of seconds after the first notification before sending a new one.
  • notification_color: There can be three types of messages green, yellow and red.
  • object_name: When system detects this object and all event filters match, it sends a notification. If you put process_statistics here, system can send you a warning, if a certain process in a pipeline took too much time.
  • save_image: System can save a reference image and send it together with a notification.
  • to_acknowledge: If turned on user needs to acknowledge this notification. If it is off, the notification will disappear automatically after some time. The time depends on the notification's color. The red ones wait longer than the green ones.


This output saves the events similar to System_DB, but uses a user-defined external DB.


The video output module saves the processed video stream as an .mp4 file. If the osd switcher is on, the detected objects, their bounding boxes and other visual information will be added to the video.

notification output


System is not supposed to save videos to files via Video output on constant basis, as BitRefine platform is not designed to work as a regular video recorder.


This output sends POST requests with JSON body to an external system.


BitRefine offers non-standard outputs that allow clients to integrate the recognition platform into a 3rd party system. Please contact BitRefine for the details.