AI Products / Services > AI Software

ZIA™ Classifier

header image

Overview

Software for recognition and classification utilizing AI and computer vision technologies, which enables an efficient classification of a large amount of image data according to the purpose such as condition observation, crime prevention, and accident prevention. Furthermore, it can perform highly accurate classification by linking with sensor data.

Example of application: Drive recorder video analysis

Drive recorder original video source : JAF MEDIAWORKS Co.,Ltd.https://www.drive-drive.jp/

Features

Optimal classification specialized for required specifications and usage environment
Classification according to the specifications required by customers is conducted by combining the image processing technology that the Company has cultivated so far and its proprietary AI technology. Efficient and highly precise classification is possible through optimizing according to the usage environment.
Pre-trained models available
A variety of pre-trained AI models, such as object detection, recognition, and tracking, are available and can be quickly applied to customer's systems.
Improved classification precision
The engine optimized for the required specifications eliminates the dispersion of judgment by humans and provides highly precise classification results.
Furthermore, linking with sensor data is also possible, enabling more detailed classification. Also, even images difficult to be classified by image processing technology can be appropriately classified by AI technology.

Recommended Usage Environment

  On-premises / Cloud
Operating environment
  • Ubuntu 18.04 (x86_64) installed
  • Main memory: 8GB or more recommended
  • NVIDIA GPU that can use CUDA 10.0 (Video memory 6GB or more recommended)

Implementation Steps

Customer's product development process
1. Survey and Planning
  • Propose the optimal AI utilization plan for the customer's request
  • Propose solutions to tasks from both software and hardware sides for PoC development
2. PoC Development
  • Collect actual data from customers, or collect and create necessary data itself, performs annotations, and create learning model
  • Evaluate the performance and recognition accuracy of the created learning model
    Achieve early PoC development by utilizing proven trained models
  • Analyze, and examine and formulate classification conditions
  • The hardware platform for PoC development can run on a GPU-equipped PC, on-premises server, or cloud environment
3. System Development / Proof of Service
  • Select algorithm and optimize learning model design according to customer's requirements such as performance, recognition accuracy and classification conditions
  • Analyze and set classification conditions for each customer
  • Connect to the customer's system and perform proof of service
4. Operation after Product Implementation
  • Regular re-training and optimization possible
  • It is possible to update the trained model according to changes in customer's requirements on recognition accuracy and performance

Related News

expand_less