1
00:00:00,450 --> 00:00:04,925
On a busy production line, a single camera can show only part of the picture.

2
00:00:07,962 --> 00:00:09,025
The front looks normal.

3
00:00:09,338 --> 00:00:10,937
A warning sits on the other side.

4
00:00:14,450 --> 00:00:18,362
A blocked view creates a different problem: the evidence is missing.

5
00:00:20,950 --> 00:00:22,763
Meet VisualGuard, by YOUNG-FU.

6
00:00:23,075 --> 00:00:23,775
Two views.

7
00:00:24,100 --> 00:00:26,013
One clear next step for the operator.

8
00:00:28,475 --> 00:00:32,088
Here is the working application, using paired synthetic images.

9
00:00:32,400 --> 00:00:35,900
OpenCV checks image quality and red candidates.

10
00:00:36,212 --> 00:00:40,850
Usable views with no red candidates lead to a recommendation for human review.

11
00:00:42,408 --> 00:00:44,246
Add a red indicator in the side view.

12
00:00:44,558 --> 00:00:50,196
VisualGuard changes the recommendation to hold automation, and shows the visual evidence behind it.

13
00:00:52,421 --> 00:00:56,346
Block the front view, and the result changes to request new images.

14
00:00:56,671 --> 00:01:00,221
The system distinguishes an alarm from evidence it cannot use.

15
00:01:02,408 --> 00:01:04,171
The operator supplies a new pair.

16
00:01:04,483 --> 00:01:07,171
Inspection runs again and returns to review.

17
00:01:07,496 --> 00:01:11,471
People stay in charge; this prototype sends no machine commands.

18
00:01:13,408 --> 00:01:19,471
The value is a usable next step: review the evidence, recommend a hold, or request another view.

19
00:01:19,783 --> 00:01:22,433
Each recommendation includes its supporting trace.

20
00:01:24,421 --> 00:01:29,171
The same OpenCV inspection core runs locally and on AWS Lambda.

21
00:01:29,483 --> 00:01:33,671
ECR stores the container, and CloudWatch records cloud execution.

22
00:01:35,288 --> 00:01:41,138
Four saved AWS calls demonstrate health, alarm, clear, and missing-view behavior.

23
00:01:41,462 --> 00:01:45,512
The review viewer opens those records without requiring cloud credentials.

24
00:01:47,033 --> 00:01:53,458
Across five hundred forty synthetic pairs, exact decision agreement was ninety-four point two six percent.

25
00:01:53,771 --> 00:01:56,508
All thirty round red stickers caused false stops.

26
00:01:59,108 --> 00:02:05,958
Fifteen archived exhibition images produced no red candidates, including a panel with a visible red lamp.

27
00:02:06,271 --> 00:02:09,596
Real-image detection remains the next engineering challenge.

28
00:02:11,158 --> 00:02:11,896
VisualGuard.

29
00:02:12,246 --> 00:02:14,558
From two views to an informed next step.

30
00:02:14,883 --> 00:02:18,671
Built by YOUNG-FU with OpenCV five and AWS.