Aiming for a seminar from the perspective of field engineers

Problem solving using image analysis technology is progressing in various industries, such as appearance inspection in the manufacturing industry, automatic driving in robots, and management of product shelves and staff in retail. The number of on-site engineers in each industry who support the technology has increased significantly, first touching the NVIDIA Jetson developer kit and proceeding with in-house development.

Therefore, we had to learn how to understand the tools (SDK) used in image analysis application development and how to use open source software (OSS) such as object detection (YOLO). There were many things. We have received many feedback that if you start by self-study, you will get stuck in a number of pitfalls and you will not be able to move forward.

There is a lot of information in the world that can be used as a reference for image analysis application development. I just want to break it down and deliver information that can avoid pitfalls in advance.

So Macnica engineers spent several months developing a lost item detection system. Specifically, it is an application that combines YOLOv4, which can detect objects and people, with OpenCV, which can calculate distance, transfers information via MQTT, and displays the results of lost item detection on Node-LED. This development was carried out by an engineer in his third year at the company, who borrowed knowledge from within the company and struggled with it as it was his first time. Based on the knowledge he gained, the seminar content included points that people might have trouble with and points that he himself had difficulty with.

3-day intensive course to learn the development flow of AI image analysis

It is composed of algorithm selection, Jetson implementation, and deployment so that you can learn the entire development flow. In the background, we referred to the feedback from customers at the previous seminar held in December 2020.

"I would like to know the series of development flow for creating an image analysis application on the blank NVIDIA Jetson."

"I want to know the whole picture of system development"

It's true that we've held articles and seminars that provide in-depth explanations of specific parts of development, but we've learned from your valuable feedback that there weren't enough to grasp the overall picture. Therefore, the engineering team gathered and realized a configuration that allows you to learn the whole process from introduction to implementation and deployment in edge computing development.

Agenda

Agenda

Day1: Algorithm selection

There is a lot of information about object detection YOLO on the internet, but how can we learn from the papers and how to choose? Aldrin from Macnica AI Research & Innovation Hub will explain in detail. Reducing false positives is the key to fully automating the process of detecting forgotten items, which has traditionally been managed manually using surveillance cameras. We introduced examples of introducing AI to conventional approaches such as background subtraction and utilizing other computer vision technologies (YOLO + OpenCV).

Day2: Jetson implementation

The core of this seminar series, from system development to operation confirmation with Jetson Xavier NX, is explained by an engineer who actually took on the challenge of system development. At the beginning, we demonstrated the complete left-behind property detection system, and showed how it can be detected in situations when you forget your bag and mobile phone. "What you can learn on Day 2 is

・Jetson implementation method of the lost property detection system

・Environmental information necessary for OSS (lost property detection) development

・Data processing method using MQTT and Node RED

is! ”, and the enthusiastic presentation began, and 277 participants listened to the end.

Day3: Deployment

Mr. Furuse, a veteran engineer, begins by talking about how difficult it is to install a developed application. For applications that use open source libraries, it is necessary to consider dependencies with a large number of libraries, and doing this manually is prone to errors. Therefore, I explained it in order by dropping it into a very detailed procedure so that I can containerize it with Docker and automate complicated installation work and model conversion.

Sample code is posted on GitHub

Details of the applications introduced in this seminar are posted on the GitHub repository.
https://github.com/MACNICA-CLAVIS-NV/abandoned_object_detection

You can download the sample code from the link.

A lively Q&A session and carefully selected Q&A are now available

During the 20-30 minute question and answer time on Days 2 and 3, we received a total of 71 questions from the participants. There are many deep technical questions that come up because the development has actually started, and we can feel that the use of the Jetson developer kit is spreading, and we share a Q&A page that you can look back on when you get stuck. I want to publish it in a separate article. Please refer to it.

Implementation and Deployment of NVIDIA Jetson Image Analysis App Hands-on Q&A Summary

Easy registration to watch on-demand videos

If you register for the 3-day online seminar that ended in great success using the form below, we will send you a URL where you can watch the on-demand video. If you missed it or are a participant and want to review it, please take this opportunity to watch it!