1.Introduction

This manual explains how to register the AI On-site Learning application (hereinafter referred to as "the product") to cameras, how to make settings before starting operation, and how to perform operations. The camera must be registered before use. Be sure to read the camera's instruction manual when using the camera. The settings shown in this manual may differ depending on the camera model used and the i-PRO Setting Tool version (hereinafter referred to as iCT).

1.1Product Overview

This software allows you to easily operate AI network cameras (sold separately, hereinafter referred to as "cameras"). It allows you to add objects that can be detected by AI Video Motion Detection installed on the cameras. In addition, if you cannot obtain the expected detection accuracy, the camera can learn scenes that suffer from problems in accuracy and improve the detection accuracy of AI Video Motion Detection.

For more information on AI Video Motion Detection, see the following URL.
→AI Video Motion Detection

Use the iCT to set up this product.

・For more information on the iCT, see the following URL. Use iCT V3.10 or later to set up this product.

→iCT

・See the following URL for more information on compatible models and software versions of this product.

→Technical information
<C0103>

* This is the control number. Enter this number to search for information on our support website.

This product is required for each camera.

1.2Features

Terms appearing in this product description are described below.

Learning images

Shows the still image to be used for learning. Use the images directly shot from the camera or JPEG/PNG images saved on your PC. These are saved in the camera.

Detector

This is the learning result created by AI from multiple learning images. These are saved in the camera.

・

The following functions improve the accuracy of AI Video Motion Detection.

-

Adding a new detection object: Learn and monitor objects such as forklifts other than human, vehicle, and bicycle that are detect by AI Video Motoin Detection.

-

Improving false detection: Learn about scenes that falsely detect objects other than human, vehicle and bicycle as human, vehicle and bicycle in AI Video Motion Detection and improve false detection.

-

Improving missed detection: Learn scenes where moving objects (human, vehicle, and bicycle) are difficult to detect by AI Video Motoin Detection and improves missed detection.

・

You can define and learn up to five detection objects for each detector.

The forklift was not detected prior to learning, but [Adding a new detection object] function enables the forklift to be detected after learning, and the alarm is activated when the forklift enters the detection area.

1.3Camera Installation Conditions

・

The following shows the minimum and maximum detection sizes for the product.


Network Camera Resolution	Minimum Detection Size (Pixels)	Maximum Detection Size (Pixels)
2 MP	100 x 50 or 50 x 100	(W)960 x (H)540
5 MP	133 x 66 or 66 x 133	(W)1280 x (H)960
6 MP	173 x 87 or 87 x 173	(W)1664 x (H)936
8 MP	200 x 100 or 100 x 200	(W)1920 x (H)1080

・

Make sure other environmental recommendations align with the recommended environment for expansion software used with this product.

For more information on AI motion detection applications, see the following URL.
→AI Video Motion Detection

1.4Precautions

The following events may cause missed or false detections.

・

When there is only a small difference in brightness between the background and the subject.

・

When the image brightness is low, such as at night.

・

When the subject is out of focus.

・

When the subject is blurred.

・

When the subject is too small or too large.

・

When the subject is overexposed or too deep in shadow.

・

When lighting conditions change easily, such as outdoors or near windows.

・

When there is interference from external light, such as sunlight or car headlights, or reflected light, such as reflections from puddles of water.

・

When strong external light shines in, casting shadows on the subject or other objects.

・

When the camera lens is wet or dirty.

・

When the camera shakes.

Other precautions are described below.

・

If you change the camera image settings before or after learning, it may cause a failure or false detections. Perform the following procedures again when you change the camera's [Image capture mode] or [Image rotation] settings.

→3.2　Saving the Learning Image

→3.3　Set bounding boxes and learn

・

[Image rotation] settings supported by this product are [0° (off)] and [180° (Upside/down)].

・

You need to collect and train with images that show the expected brightness levels during operation.

・

When using under IR-LED, you need to learn with images under IR-LED.

・

The detection performance of on-site learning function depends on the learning results and the environments you use. Not suitable for use in use cases where perfect detection performance is required. Be sure to check the operation at the installation site. We do not guarantee any detection performance.

・

This product cannot be set or perform learning from multiple PCs at the same time.

・

When the camera7s [Configuration data initialization] is executed, the camera settings, learning images, and detectors are not initialized. Note, however, that this may be affected by the image quality settings on the camera, so be sure to check the operation again.

・

If this product is uninstalled, the learning images and detectors will be deleted.