7.Settings in environment without camera connection

This chapter describes the procedure for setting the product in the camera disconnection environment.
Check the settings from the table below and refer to the appropriate chapter.
Setup method
Reference
[Image acquisition] is performed in the camera connection environment, and [AI On-site Learning setting] is performed in the camera unconnected environment (office, home, etc.)
7.1　Perform some processes in non-camera environment
[Image acquisition] and [Bounding box setting] are done in the camera-connected environment, and the rest of the process is done in the non-camera-connected environment.
7.2　Stop the work in a camera-connected environment and continue in a non-camera environment

7.1Perform some processes in non-camera environment
1.
Select [AI On-site Learning Management] from the submenu
2.
Check the check box of the camera to set the product.
This product is set by one camera. Check only one check box on the camera.
3.
Select [Perform some processes in non-camera environment]

7.1.1Image acquisition
1.
Check the checkbox for the cameras you want to collect and press [Display].
-
[Image acquisition screen] is displayed on a separate screen.
2.
Set the saved folder
-
When the Folder button is pressed, the Browse Folder screen is displayed on the separate screen.
3.
Select the destination folder for saving images and press [OK].
4.
Save the image used for learning
-
[Manual save] allows you to manually save still images each time.
-
In [Auto save camera image], when you specify [Save interval] and [Number of save images], the camera automatically saves one still image for [Number of save images] at the specified [Save interval].
・
[Manual save]
-
When [Snapshot] is pressed, one still image is saved in the save folder.
・
[Auto save]
-
[Image saving interval]: Select from [10 s], [20 s], [30 s], [40 s], [50 s], [1min], [5min], [10min], [15min], [30min], and [60min].
-
[Number of save images]: Select from [10], [20],..., [90], [100], [150], and [200].
-
[Start]: Start auto saving.
-
To stop auto saving, click [Stop].
-
During auto save, the progress is displayed beside the [Stop].

・The number of saved images is displayed at the bottom of the screen as "Number of saved images: x".


・The camera's image capture mode should be set to [16:9 mode].
5.
Press [Close] to exit [Image acquisition].

7.1.2AI On-side Learning setting
Set AI On-site Learning in the camera unconnected environment. When [Display] is pressed, [AI On-site Learning Setup screen] is displayed in a separate screen.
Detector 1 is set in the [AI On-site Learning Settings] window.

7.1.2.1Image input
Drag and drop images or folders acquired in the field as described in 7.1.1 into the dotted line frame at the top of the screen. The entered images are added to the list.
To delete an added image from the list, select the check box at the bottom left of the thumbnail of the target image and click [Delete image]. When the learned image is deleted, "No Image" is displayed on the thumbnail.

7.1.2.2Detected object setting
When [Display] is pressed, [Detection object setting screen] is displayed in a separate screen.
In [Set Detection Object], configure the following settings for up to five detection objects.
1.
[Function]
　
You can set the Detection Function for each of the five [Detection Objects].
　
[New object]: You can add a new object you want to detect.
　
[Improving false detection]: Improve false detection of human, vehicle and bicycle.
　
[Improving missed detection]: Improve missed detection of human, vehicle and bicycle.
2.
[Name]
　
Set the name of detection object. You can set up to 20 half-pitch characters.
3.
[Target]
　
If [Improving false detection] or [Improving missed detection] is selected for [Function], select detection object to be improved.
　
Select from [Human], [Vehicle], and [Bicycle].
4.
Trash box icon
　
If you press the [Set] after clicking, the settings for the selected detection object will be deleted.
　
Settings to be removed: [Function], [Name], [Target], Settings of bounding box, Settings of [Training data expansion]
5.
[Set]
　
Click to reflect the settings.

7.1.2.3Bounding box setting
Click the image displayed in the image list to display [Learning Frame Setting Screen] on the separate screen.
1
Select [Add Bounding box] in [Mode] to draw a bounding box on the image.
-
You can set the learning frame as a rectangle by dragging it on the image.
-
Right-clicking inside a learning frame brings up a menu in the "Learning Frame Settings" window. Select [Delete] in the menu to delete the right-clicked bounding box. Select [Change to (the name of  detected object)] in the menu to change the bounding box to that of the selected detected object. The same operation can be performed when [Edit Bounding box] is selected in [Mode].
2.
Select [Edit Bounding box] in [Mode] to edit a bounding box on the image.
-
Clicking on a drawn bounding box allows you to change the size and position, or delete it by clicking on the X icon.
-
Clicking the Icons for the detection results (gray frames) of human/vehicle/bicycle/learned objects will set the selected detection object to the bounding box.
-
The gray frames of human/vehicle/bicycle/learned objects can be shown or hidden with the "Show/hide the gray frame" checkbox.

・If the size of the learning frame you draw is too small or too large, the size is automatically adjusted. See the following for the minimum and maximum sizes.


→1.3　Camera Installation Conditions

・The detection results for human/vehicle/bicycle/learned objects are displayed when the [Show the detection frame] checkbox is checked in the [AI On-Site Learning settings screen].


・You can set up to 100 bounding box per image.
3.
Press [Next] to set the bounding box for the remaining images.
-
The setting screen for the next image is displayed.
4.
When you have set all the learning images, click [Set].
-
[Learning Frame Setting Screen] closes, and you are returned to the [AI On-site Learning setting screen].

7.1.2.4Execute learning
1.
Select the image you want to use for learning.
-
Select the check box at the bottom left of the thumbnails of the target images as shown.
-
Learning requires at least 10 images.You can learn up to 200 images.
-
The maximum number of bounding boxes that can be included in learning images are 1000.

・When bounding boxes are set, the check boxes at the bottom left of the thumbnails are automatically checked.


・When the [Show the detection frame] is checked, the detection results for human/vehicle/bicycle/learned objects are displayed in a gray frame. Detection results for human/vehicle/bicycle/learned objects that exceed the detection thresholds will be displayed. For more information on detection thresholds, please refer to the following.


→7.2.2.5　Confirmation of detection results
2.
Press [Start Learning] to execute learning.
-
When learning is completed, the confirmation screen is displayed.

7.1.2.5Confirmation of detection results
Confirm the learning results. Press [Display] to open the [Detection Result Confirmation window] on the separate window.
In the [Detection result confirmation screen], enter the image for which you wish to simulate detection accuracy, and the detection results will be displayed. While viewing the displayed detection results, change the threshold value and adjust it so that the detection frame is displayed correctly.
1.
[Human identification threshold]
-
Set the detection threshold for people. The smaller the value, the easier it is to detect people, but it is also likely to return more false detections.
10 to 99
Default: 20
2.
[Vehicle identification threshold]
-
Set the detection threshold for vehicles. The smaller the value, the easier it is to detect vehicles, but it is also likely to return more false detections.
10 to 99
Default: 70
3.
[Bicycle identification threshold]
-
Set the detection threshold for bicycles. The smaller the value, the easier it is to detect bicycles, but it is also likely to return more false detections.
10~99
Default: 55
4.
[Threshold setting] for new objects
-
Set the detection threshold for objects 1 to 5. The smaller the value, the easier it is to detect objects 1 to 5, but it is also likely to return more false detections.
10 to 99
Default: 50
5.
[Hide icons]
-
Select to hide the detection object icons.
The following formats are supported for input images: .jpg and .png.

7.1.2.6Saving training data
Saves the AI On-site Learning settings as training data.
1.
Press [Save].
-
The folder browsing screen will appear in a separate window.
2.
Select a folder to save the training data, and press [OK].
-
Saving of the training data will start.
-
The saved training data will be saved as a zip file.
3.
Press [Completed] to exit AI On-site Learning setting.
To clear the training information of detector and execute a new training, press the [Initialize] button. Note that pressing the [Initialize] button will delete all values and images that have already been set in the AI On-site Learning settings.

7.1.3Upload and configuration of detection results
Upload the training data saved in [AI On-Site Learning setting] to the camera.
1.
Press [Start].
-
The file selection screen will appear in a separate window.
2.
Select the training data (zip file) saved in 7.1.2.6 and press [Open].
-
A confirmation screen will appear on a separate screen. Press [OK] to start uploading.
3.
When the upload is complete, press [Display].
-
The demo screen will appear in a separate window. See below for details on the demo screen.
→5.　Demo Screen
If the check box for the target camera for which you want to upload data is not checked, you will not be able to press [Start]. Please check only one check box for the target camera and press [Start].

7.2Stop the work in a camera-connected environment and continue in a non-camera environment
1.
Close [AI On-site Learning setting screen].
-
Return to the [AI On-site Learning Management] screen.
2.
Select [AI On-site Learning Management] from the sub-menu.
3.
Check the checkbox of the cameras for which you want to configure for this product.
Configure this product one camera at a time. Please check only one check box for each camera.
4.
Click [If you stop the work and continue in a non-camera environment].
-
The item will expand and the settings will be displayed.

7.2.1Download
Download the data (learning images, setting of bounding boxes) from the camera in order to bring the data in the process of work back to the non-camera environment.
1.
Press [Start].
-
The folder browse screen will appear on a separate screen.
2.
Select the folder where the data will be saved, and press [OK].
-
Data downloading starts.
-
The downloaded data will be saved as a zip file.
If the check box for the target camera for which you want to download data is not checked, you will not be able to press [Start]. Please check only one checkbox for the target camera before pressing [Start].

7.2.2AI On-site Learning setting
In [AI On-site Learning setting], you can set up AI On-site Learning in an environment where the camera is not connected. Click [Display] to display the [AI On-site Learning setting screen] in a separate window.
In the [AI On-site Learning settings screen], you can configure the settings for detector 1.

7.2.2.1Enter the downloaded file
Drag and drop the zip file or image downloaded from the camera in 7.2.1 into the dotted frame at the top of the screen. The entered image will be added to the list.
If you want to remove the added image from the list, check the checkbox at the bottom left of the thumbnail of the target image and press the [Delete Image].

7.2.2.2Detected object setting
Press [Display] to display the [Detected object setting screen] in a separate window.
In [Detected object setting screen], configure the following settings for up to five detection objects.
1.
[Function]
　
You can set the Detection Function for each of the five [Detection Objects].
　
[New object]: You can add a new object you want to detect.
　
[Improving false detection]: Improve false detection of human, vehicle and bicycle.
　
[Improving missed detection]: Improve missed detection of human, vehicle and bicycle.
2.
[Name]
　
Set the name of detection object. You can set up to 20 half-pitch characters.
3.
[Target]
　
If [Improving false detection] or [Improving missed detection] is selected for [Function], select detection object to be improved.
　
Select from [Human], [Vehicle], and [Bicycle].
4.
Trash box icon
　
If you press the [Set] after clicking, the settings for the selected detection object will be deleted.
　
Settings to be removed: [Function], [Name], [Target], Settings of bounding box, Settings of [Training data expansion]
5.
[Set]
　
Click to reflect the settings.

7.2.2.3Bounding box setting
Click the image displayed in the image list to display [Learning Frame Setting Screen] on the separate screen.
1.
Select [Add Bounding box] in [Mode] to draw a bounding box on the image.
-
You can set the learning frame as a rectangle by dragging it on the image.
-
Right-clicking inside a learning frame brings up a menu in the "Learning Frame Settings" window. Select [Delete] in the menu to delete the right-clicked bounding box. Select [Change to (the name of  detected object)] in the menu to change the bounding box to that of the selected detected object. The same operation can be performed when [Edit Bounding box] is selected in [Mode].
2.
Select [Edit Bounding box] in [Mode] to edit a bounding box on the image.
-
Clicking on a drawn bounding box allows you to change the size and position, or delete it by clicking on the X icon.
-
Clicking the Icons for the detection results (gray frames) of human/vehicle/bicycle/learned objects will set the selected detection object to the bounding box.
-
The gray frames of human/vehicle/bicycle/learned objects can be shown or hidden with the "Show/hide the gray frame" checkbox.

・If the size of the learning frame you draw is too small or too large, the size is automatically adjusted. See the following for the minimum and maximum sizes.


→1.3　Camera Installation Conditions

・The detection results for human/vehicle/bicycle/learned objects are displayed when the [Show the detection frame] checkbox is checked in the [AI On-Site Learning settings screen].


・You can set up to 100 bounding box per image. A total of up to 1000 learning images can be set.
3.
Press [Next] to set the bounding box for the remaining images.
-
The setting screen for the next image is displayed.
4.
When you have set all the learning images, click [Set].
-
[Learning Frame Setting Screen] closes, and you are returned to the [AI On-site Learning setting screen].

7.2.2.4Execute learning
1.
Select the image you want to use for learning.
-
Select the check box at the bottom left of the thumbnails of the target images as shown.
-
Learning requires at least 10 images.You can learn up to 200 images.
-
The maximum number of bounding boxes that can be included in learning images are 1000.
When bounding boxes are set, the check boxes at the bottom left of the thumbnails are automatically checked.
2.
Press [Start Learning] to execute learning.
-
When learning is completed, the confirmation screen is displayed.

7.2.2.5Confirmation of detection results
Confirm the learning results. Press [Display] to open the [Detection Result Confirmation window] on the separate window.
In the [Detection result confirmation screen], enter the image for which you wish to simulate detection accuracy, and the detection results will be displayed. While viewing the displayed detection results, change the threshold value and adjust it so that the detection frame is displayed correctly.
1.
[Human identification threshold]
-
Set the detection threshold for people. The smaller the value, the easier it is to detect people, but it is also likely to return more false detections.
10 to 99
Default: 20
2.
[Vehicle identification threshold]
-
Set the detection threshold for vehicles. The smaller the value, the easier it is to detect vehicles, but it is also likely to return more false detections.
10 to 99
Default: 70
3.
[Bicycle identification threshold]
-
Set the detection threshold for bicycles. The smaller the value, the easier it is to detect bicycles, but it is also likely to return more false detections.
10~99
Default: 55
4.
[Threshold setting] for new objects
-
Set the detection threshold for objects 1 to 5. The smaller the value, the easier it is to detect objects 1 to 5, but it is also likely to return more false detections.
10 to 99
Default: 50
5.
[Hide icons]
-
Select to hide the detection object icons.
The following formats are supported for input images: .jpg and .png.

7.2.2.6Saving training data
Saves the AI On-site Learning settings as training data.
1.
Press [Save].
-
The folder browsing screen will appear in a separate window.
2.
Select a folder to save the training data, and press [OK].
-
Saving of the training data will start.
-
The saved training data will be saved as a zip file.
3.
Press [Completed] to exit AI On-site Learning setting.
To clear the training information of detector 1 and execute a new training, press the [Initialize] button. Note that pressing the [Initialize] button will delete all values and images that have already been set in the AI On-site Learning settings. However, the training information except for detector 1 will not be cleared.

7.2.3Upload and configuration of detection results
Please see below.
→7.1.3　Upload and configuration of detection results