Documentation for Axxon One 2.0. Documentation for other versions of Axxon One is available too.

Previous page Hardware requirements for neural analytics operation  Optimizing the operation of neural analytics on GPU in Windows OS Next page

Description

To train neural networks, you must collect and submit to AxxonSoft videos and images from your actual cameras taken in the same resolution and in the same conditions as in your future application.

For example, if you need your neural network to work outdoors, videos must contain all weather conditions (sun, rain, snow, fog, and so on) at different times of day (morning, afternoon, evening, and night).

If the collection requirements for the data submitted for training the neural network model are met, and if you operate the neural network in conditions that are as similar as possible to the conditions in which the data was collected, we guarantee the overall accuracy (training in operating conditions) of neural network analytics from 90% to 97% and the percentage of false positives of 5-7%. For standard neural networks (training wasn't performed in operating conditions), we guarantee the overall accuracy of 80-95% and the percentage of false positives of 5-20%. The standard neural networks are located in the C:\Program Files\Common Files\AxxonSoft\DetectorPack\NeuroSDK directory.

The requirements can be changed or added to at any time

General requirements for collected data
  • When collecting data, specific requirements for object images, scene, angle, lighting, and video stream must be met for those detectors that you plan to use (see Detectors).

  • If it is required to train a neural network in different conditions of time of day, lighting, angle, object types, or weather, you must collect video data in equal proportions for each condition, that is, it must be balanced.

Example. It is necessary to detect a person in the surveillance area at night and during the day.

  • Data collected correctly:
    • four videos of the surveillance area, each five minutes long are submitted for training;
    • the object of interest appears in the frame in each video;
    • two videos are recorded in nighttime conditions, two—in daytime conditions.
  • Data collected incorrectly:
    • three videos of the surveillance area, each five minutes long are submitted for training;
    • the object of interest appears in the frame in each video;
    • two fragments are recorded in nighttime conditions, one—in daytime conditions
Additional requirements for collected data for each neural analytics tool
 
DetectorRequirements
Neural filterAt least 1000 frames containing different objects of interest in the given scene conditions, and the same number of frames containing no objects (noise frames)
Neural tracker3–5 minutes of videos containing objects of interest in the given scene conditions. The greater the number and variability of situations in the scene, the better
Stopped object detector
Human pose detector

10 second video of a scene with no persons.
At least 100 different persons in the given scene conditions.
Attention! Different conditions mean, among others, different positions of a person in a scene (bending, different positions of body parts, and so on)

Equipment detector

A list of all reference equipment with examples must be collected from the object and coordinated with the analytics manufacturer (see Example of providing a list of valid equipment at the facility for the Equipment detector).
3-5 minutes of videos with personnel in the given scene conditions.
Personnel must move and change positions in the collected videos, as well as remove and put on equipment at intervals of 30 seconds.
Since the equipment detectors are designed for artificial constant lighting, videos in other lighting conditions aren't required

Equipment detector VL
Fire detectorAt least 1000 frames containing different objects of interest in the given scene conditions, and the same number of frames containing no objects (noise frames)
Smoke detector
Neural classifierAt least 1000 frames containing different objects of interest in the given scene conditions, and the same number of frames containing no objects (noise frames)
  • No labels