NEWS LETTER

 By SEYEONTECH

Examples of customer LPR development using the Open AI SDK

SeyeonTech supports an open AI SDK to help customers build their own AI IP cameras. By utilizing the open AI SDK, customers can utilize their own AI learning models by porting them to IP cameras. In general, unless you are an IP camera manufacturer, it has been impossible to mount your own AI technology on an IP camera.

SeyeonTech supports a PC learning environment and a standardized conversion tool for porting AI learning models to IP cameras. The conversion tool is implemented on the web, so customers can check the results after porting AI learning models to IP cameras by themselves. The advantage is that customers do not need to share their DB and learning models with Seyeon Tech. If AI technology is needed at the edge, it can be implemented on the camera using the 'Open AI (Artificial Intelligence) SDK'.

Table of Contents : 

1. Customers who have commercialized LPR by applying the 'Open AI SDK'

2. Types of LPR-enabled license plates developed by customers 

Overview : 

1. Customers who commercialized LPR by applying 'Open AI SDK' 

The customer who commercialized LPR this time is a company that has been operating LPR-related business for a long time in PC and embedded environments. They have already implemented LPR on PC, Raspberry Pi, Jetson, etc. and have various kinds of license plate DB. They commercialized LPR after repeated testing and performance verification through an 'open AI (Artificial Intelligence) SDK'. Multiple learning models were used to improve LPR accuracy, and Seyeon Tech supported the correlation between learning models and some post-processing. The post-processing to improve LPR performance was done in parallel on a remote collection server.

For porting, we trained using MobileNet-SSD's detection model in the PyTorch framework environment. In addition, it is necessary to adjust hyper parameters within 20 such as epoch, batch size, learning rate, and anchor box to increase the learning recognition rate. After completing this process, the AI model that can be ported to the Seyeon Tech AI IP camera is ready and can be ported to the camera according to the SDK order. Camera metadata transmission, PC Bounding Box display, and class information notation are included in the SDK, and porting is finalized by modifying the class information. Customers do not need to share their labeled DB and AI learning models with SeyeonTech, and can apply them to IP cameras by themselves.

2. customer-developed LPR-enabled license plate types 

Customer LPR models can be ported to any CeylonTech 2M to 12M camera. Currently, it is optimized for 2M cameras and modules, and supports recognition speed around 10 FPS. We are preparing to support LPR for 5M to 4K cameras in Q1 24 with no speed difference. The LPR feature is licensed separately and is currently available for 2M varifocal bullet cameras.

Other 

The following AI learning models are in progress at each partner company.

- Development of flame detection and smoke detection models

- Development of tolerance and VMS integration

- Fall, assault learning and alarm system application

- Development and operation of a 4-lane license plate detection system based on a 4K global shutter camera

- Detecting people in the robot arm operation area and interlocking alarm equipment

SeyeonTech aims to provide an 'open AI IP camera development platform' to development-based customers who need to utilize AI technology. We want to increase the utilization of IP cameras while sharing core technologies with partners.