Documentation for DetectorPack PSIM 1.0.1.

Previous page General information on metadata  Data collection requirements for neural network training Next page

Requirements for software and hardware platform

DetectorPack PSIM uses neural network analytics. The following features are available on the basis of neural networks:

  1. Neurotracker (see Neurotracker).
    Neurotracker detects only objects of a specified class using a neural network. The Neurotracker is more accurate than the regular one, and detects even static objects, but it requires more computing resources.

  2. Fire and Smoke detection (see Fire detection and Smoke detection).
    Neural network detects fire and smoke in a frame.

  3. Person location tracker (see Person location tracker). 
    Neural network determines each person's skeleton and detects poses that can represent a security threat.
  4. Neurocounter (see Neurocounter).
    Neural network counts the number of objects in a specified area.
  5. Sweethearting detection (see Sweethearting at checkout detection).
    Neural network detects the theft of goods by cashiers who don't scan barcodes of some items at the checkout.
  6. Equipment detection (see Equipment detection (PPE)).
    Neural network divides the human body into segments, detects the equipment (PPE) on a specified body segment, and checks its condition. Segmenting and classification neural networks are used to operate the Equipment detection (PPE).

The quality of work and the resource consumption of the neural network-based detection tools directly depend on the optimization of the neural network model used.

Notes

The file names for each neural network model depend on the device on which the neural network will operate:

  • If you intend to use a CPU or Intel GPU (integrated video core), then it is necessary to train the neural network file using the OpenVINO toolkit, and the name of the neural network file must contain the _openvino substring at the end. For example, test1_openvino.ann.
  • If you intend to use an NVIDIA GPU graphics processor, then the name of the neural network file should contain only the name of the neural network. For example, test1.ann.
  • If you intend to use the Intel NCS, then the name of the neural network file must contain the _movidius substring at the end. For example, test1_movidius.ann.

Before you start configuring the neural network-based detection tools, you must contact the AxxonSoft technical support and request the model files of the trained neural networks. Technical support specialists will request the required data (see Data collection requirements for neural network training) and then provide the files for each neural network model. These files must be distributed to all Servers where you plan to use the detection tools.

Attention!

The startup (initialization) time of each neural network on NVIDIA GPU can take 2-3 minutes, depending on the neural network model that you use. Until initialization is complete, no events will be received from detection tools.

  • No labels