WEBVTT

00:04.340 --> 00:07.430
or in its threat target automated

00:07.430 --> 00:09.486
recognition or otter for short is an

00:09.486 --> 00:11.720
artificial intelligence Ai project that

00:11.720 --> 00:13.776
is developing a new machine learning

00:13.776 --> 00:15.776
capability that would be integrated

00:15.776 --> 00:17.942
into the adopt software on E . O . D .

00:17.942 --> 00:20.164
Mobile field kits . The objective is to

00:20.164 --> 00:22.164
decrease the search time associated

00:22.164 --> 00:24.109
with how an E . O . D . Technician

00:24.109 --> 00:26.164
identifies found unexploded ordnance

00:26.164 --> 00:28.220
items upon initial reconnaissance of

00:28.220 --> 00:30.331
AUX . So item E . O . D . Technicians

00:30.331 --> 00:30.320
photograph from various angles and

00:30.320 --> 00:32.376
reference images against items found

00:32.376 --> 00:34.598
within the adopt software to positively

00:34.598 --> 00:36.598
identify an item , text must search

00:36.598 --> 00:38.820
through numerous items within adopts to

00:38.820 --> 00:40.987
confirm which item may need to perform

00:40.987 --> 00:43.153
a render safe for disposal procedure .

00:43.153 --> 00:45.209
The newly added image search feature

00:45.209 --> 00:47.376
will allow you D technicians to upload

00:47.376 --> 00:47.300
a photograph from their handheld device

00:47.310 --> 00:48.866
and an ordnance search will

00:48.866 --> 00:51.088
automatically be performed based on the

00:51.088 --> 00:53.199
photograph . This drastically reduces

00:53.199 --> 00:55.254
search time and narrows the scope of

00:55.254 --> 00:57.477
possible item matches . This will allow

00:57.477 --> 00:57.370
the EU DE technician to perform their

00:57.370 --> 00:59.537
task more efficiently and with greater

00:59.537 --> 01:01.537
certainty to develop otter physical

01:01.537 --> 01:03.759
inert ordnance items were acquired from

01:03.759 --> 01:05.537
the N . S . W . C . Indian head

01:05.537 --> 01:07.759
technical library to be three D scanned

01:07.759 --> 01:09.981
and compiled into a database set . This

01:09.981 --> 01:11.870
initial database because of known

01:11.870 --> 01:14.092
ordinance items on our will rely on the

01:14.092 --> 01:16.148
input of field and fleet E . O . D .

01:16.148 --> 01:16.100
Units to expand the database in the

01:16.100 --> 01:18.100
future , technicians operating high

01:18.100 --> 01:20.044
precision three D scanners capture

01:20.044 --> 01:21.989
every detail on the surface of the

01:21.989 --> 01:24.211
ordnance item , color , shape , texture

01:24.211 --> 01:26.156
and markings are details needed to

01:26.156 --> 01:28.211
positively identify the item three D

01:28.211 --> 01:27.970
scans are performed on ordnance items

01:27.970 --> 01:30.026
large and small . The technician can

01:30.026 --> 01:32.192
lace together multiple scans necessary

01:32.192 --> 01:34.192
for larger items three D . Ordnance

01:34.192 --> 01:36.192
models are optimized and share with

01:36.192 --> 01:38.359
project partners specialize in physics

01:38.359 --> 01:40.414
based reality software creating high

01:40.414 --> 01:42.414
volumes of synthetic images . These

01:42.414 --> 01:44.581
synthetic images are used to train the

01:44.581 --> 01:46.581
otter algorithm to see and identify

01:46.581 --> 01:46.540
ordnance items and diverse lighting

01:46.540 --> 01:48.540
backgrounds and angles . The gaming

01:48.540 --> 01:50.762
engine can also simulate how ordinances

01:50.762 --> 01:52.873
naturally drop or obscured from you .

01:52.873 --> 01:54.818
The synthetic images are used by a

01:54.818 --> 01:56.984
computer scientists to develop machine

01:56.984 --> 01:59.096
learning image classifier model . The

01:59.096 --> 02:01.096
algorithm parameters are created so

02:01.096 --> 02:03.318
that the system inspects each ordinance

02:03.318 --> 02:05.429
item found within the photographs and

02:05.429 --> 02:07.651
records its features and gets converted

02:07.651 --> 02:09.707
into awaits file the weights file is

02:09.707 --> 02:09.380
the code generated by the machine

02:09.380 --> 02:11.491
learning training process that allows

02:11.491 --> 02:13.547
the software to recognize the unique

02:13.547 --> 02:15.547
features of a photographed ordnance

02:15.547 --> 02:17.602
item . After training the software ,

02:17.602 --> 02:17.310
the algorithm of waste fire all that

02:17.310 --> 02:19.366
are necessary to be implemented into

02:19.366 --> 02:21.699
the adult software . This enables the U .

02:21.699 --> 02:23.921
D . Technician out in the field to take

02:23.921 --> 02:23.800
photographs of ordinance and the

02:23.800 --> 02:26.022
software produces a result . The aid of

02:26.022 --> 02:27.689
software development team has

02:27.689 --> 02:29.911
implemented prototypes of the new image

02:29.911 --> 02:31.856
search feature enabling you detect

02:31.856 --> 02:33.967
users to photograph hornets using the

02:33.967 --> 02:35.967
mobile field kit . Although the new

02:35.967 --> 02:38.133
image search feature produces the best

02:38.133 --> 02:39.856
possible matches . E . O . D .

02:39.856 --> 02:41.856
Technicians must use their training

02:41.856 --> 02:43.800
skill and experience to positively

02:43.800 --> 02:43.480
confirmed the order this item

02:43.480 --> 02:44.170
encountered