News
We will be a company that connects the world as one.
News
News Letter
News Letter
Put your AI technology directly on IP cameras!
2023.09.21 16:51
Put your AI technology directly on IP cameras!
'Customers can apply AI Model without sharing the labeling DB'
'Apply to IP cameras online without any other help'
SeyeonTech has released an open SDK for customers to create their own AI IP cameras. By utilizing this, customers can directly upload AI models to the IP camera, check the results, and get their own AI IP camera. Unless you're an IP camera manufacturer, it's not practical to build your own AI technology into an IP camera.
SeyeonTech supports a standardized conversion tool to put AI models trained on a PC onto an IP camera. It is available online, and you can upload the AI model to the IP camera and check the result by yourself. The advantage is that you can check the results on the web without providing the customer's DB to Seyeon Tech. If you need multi-step reasoning or special applications in addition to the standardized contents, it is possible through technical cooperation with SeyeonTech.
Table of Contents :
1. What is an open SDK (FlexWATCH® OPEN SDK for AI)?
2. Actual customer development cases (license plate recognition, Parking management, driving vehicle enforcement, etc.)
1) License plate recognition (multi-stage inference)
2) 12M Fisheye camera Parking management
3) 4K Global Shutter Camera Cracking down on vehicles traveling in the 3rd lane
4) Smart signaling (crosswalk guiding lights)
Overview :
1. What is an open SDK (FlexWATCH® OPEN SDK for AI)?
This development kit is designed to make it easy for customers to put their own AI models on IP cameras. It guides you through the process of configuring a PC training environment for porting, training AI models, and porting the results to IP cameras. Here's a summary of the open SDK.
- Preparation of PC environment (Ubuntu 20.04)
- Creating a Python virtual environment
- MobilenetV2 +SSDLite training
- Converting the Trained model to an ONNX file
- Validating the Trained model
- Converting ONNX file to binary file working over FlexWATCH® camera
- Creating a compressed file (tar.gz) to upload to the FlexWATCH® camera
2. Actual customer development cases (license plate recognition, Parking management, driving vehicle enforcement, etc.)
1) License plate recognition (multi-stage inference)
- Two companies are currently developing a license plate recognition system using the open SDK. It was confirmed that 10 license plates can be recognized per second with one license plate.
- For license plate recognition, a two-stage inference is performed. The first step is to detect the license plate, and the second step is to extract the vehicle number and Korean class from the detected license plate. One customer has prepared a learning model for each type of license plate, and is performing license plate recognition with a total of 5 learning models.
- To operate the customer's learning model, SeyeonTech supports the post-processing of the extracted class, confidence, and coordinate information, as well as the transmission of metadata and display on a web browser. The post-processing process is based on our own API and ONVIF standard.
2) 12M Fisheye camera Parking management
- 12M Fisheye camera (EX1-412 module) to read whether cars are parked or not (up to 12 cars). Supported by 1 AI model.
- You can specify the parking location of the vehicle and park it in the area to read whether the car is parked or not.
- Real-world examples
3) 4K Global Shutter Camera Cracking down on vehicles traveling in the 3rd lane
- This is an example of a real-world application of a 3-lane illegal vehicle enforcement system utilizing SeyeonTech's 4K global box camera. Here, the camera is configured to detect vehicles in the 3rd lane in real time and send the license plate to the server for license plate recognition.
4) Smart signaling (crosswalk guiding lights)
- It was applied to a smart traffic light device that determines the presence of pedestrians and was utilized as a day and night pedestrian detection system.
Other
SeyeonTech aims to provide various IP camera development environments for development-based customers who want to utilize AI technology. In other words, we share the core technology of our 3rd party partners to increase the utilization of IP cameras. Simple detection and intelligent rules (such as camera auto-tracking) are supported by SeyeonTech as standard.
Contact Us
Seyeon Tech Sales Department
sales@seyeon.co.kr
이전글
다음글
