Neural Pocket ニューラルポケット株式会社




AI Implementation Technologies for Edge Devices

AI Implementation Technologies for Edge DevicesThe use of edge processing technology running AI models in camera devices allows video captured by cameras to be processed onsite in real-time, instantly providing abstract data that can be visualized, or stored and analyzed.Not only does this greatly reduce the communication load compared to conventional methods that require the uploading of raw video files to servers off premises, parallel distributed processing makes it possible to process video from thousands to tens of thousands of cameras in real time.Further, the ability to analyze data within the camera device on premise, eliminates the need for images and footage to be shared with remote off-premise servers, protecting personal information. Edge AI allows for solutions that can automatically delete footage as it analyzes and extracts insights without any communication with external servers nor human intervention revolutionizing the way we approach the protection of personal information.

Broad-ranging Object/Person Recognition Technology

In addition to maintaining a diverse range of AI libraries for object/person recognition, we use dedicated tools for accelerated learning to increase repertoire of detectable objects and cater to a broader range of applications.The combination of libraries used for inferring fashion items and personal attributes, and recognizing posture, line of sight and movements makes it possible to provide information that can be used for commercial, advertising and other such applications.The combination of facial recognition, and the detection of personal belongings and protective equipment is used to combat crime and to provide for safer working environments. This technology can also be applied to route analysis and work behavior analysis to help provide efficiency improvements in warehouses and factories.

Advanced Proprietary AI Library

A broad range of proprietary AI libraries are being developed to cater to an ever-increasing range of applications.For example, it is possible to add advanced recognition features to existing cameras through the use of depth perception in a monocular camera, and object recognition algorithms for ceiling cameras and infrared cameras.This is also used in object dimension measurements taken from images, drone self-localization, 3D motion analysis and blurring/defocusing-resistant number plate recognition technology, creating new application services unseen in the industry.

AI Development Platforms

The systemization of AI research and development, in particular, the use of proprietary annotation tools that utilize active learning and operational know-how, is fueling our ability to build new AI development frameworks, enabling accelerated reproducibility of results in solving new problems.In order to solve the myriad issues specific to edge devices that occur in remote, multiple-system and poor operating environments, we are developing and utilizing dashboards used for integrated device management, and security systems and diagnostic measures. We have also accumulated knowledge on providing support for a broad range of devices and communication environments, including everything from dedicated devices to smartphones and web browsers, all of which is being built as an integrated system.