As a developer or system integrator of AI vision applications, are you struggling to find an easy-to-use and efficient AI vision solution? Are you unsure about the return on investment of smart vision IoT? As a developer of AI vision algorithms, do you find that specific algorithms are easy to build but hard to turn into products? As a maker of smart cameras and smart vision devices, is your product slow to reach the market because there is no complete, easy-to-integrate system platform?
BrainFrame, an industry-first real-time video/vision AI analytics platform built on edge computing, has officially launched! In early 2020 we co-released the BrainFrame + OpenVisionCapsules open-source vision algorithm capsule solution with OpenCV.org. Users and developers can now download it for free: https://dilililabs.com/zh/download/.
The platform can be used in all kinds of continuous monitoring, tracking and statistical analysis scenarios, for example: recognizing and tracking people and vehicles; behavior analysis; item classification and production-line quality inspection; and anomaly alerts. Users or system integrators can drag and drop algorithm capsules onto the BrainFrame server and, after simple configuration, apply vision AI to many applications: industry, retail, administration, transportation, healthcare and more. By providing a powerful real-time vision platform for computer vision and neural networks that is easy to integrate, extend and deploy, it lets system integrators and developers handle all kinds of intelligent video analytics tasks with ease.
BrainFrame has the following features:
-
Edge computing, ready to use as soon as it is installed: 1) Video in, structured data out. Video processing and AI inference run on edge computing devices, extracting real-time insights from fixed-camera video streams to help staff monitor and track what happens on site, or to record data. Structured data is output as real-time alerts, or as reports for later statistical analysis and decision-making. 2) The system comes with many vision applications*: queue management and seat occupancy/attendance in all kinds of venues, shelf space and inventory management, safe-production compliance, restaurant/retail staff management, bank service and lobby management, traffic/urban management and analysis, and more — no extra development needed. 3) Through a graphical interface, staff set AI rules for different areas of the video according to business needs, for monitoring, tracking or real-time alerts. Alert methods and report formats can also be adjusted to business needs. Getting Started - BrainFrame Documentation.
-
Easy to deploy, flexible to extend: 1) The system has a user-oriented graphical interface that is simple and easy to use — a What It Sees Is What You Get (WISIWYG) smart vision platform. Download the BrainFrame server software and install it on Linux, and it is ready to use with no programming. Following the instructions, installing the software on an x86 computer usually takes about an hour: https://dilililabs.com/docs/user_guide/server_setup/. The system comes with a graphical client for Linux or Windows, including database/dashboard charts and a real-time alert interface. Enter the IP address of an ordinary camera directly in the client and you can start intelligent video operations right away. Configuring the system takes only a few minutes: https://dilililabs.com/docs/user_guide/client_setup/. 2) If the computer has a GPU, you can use any of the released high-performance, high-accuracy algorithm capsules; without a GPU, you can use cost-effective OpenVino-compatible algorithm capsules. 3) BrainFrame can run as a standalone AI video application managing a few cameras, or as an enterprise or industrial AI video service that uses clustered computing* to connect multiple sites and support thousands or even millions of camera streams.
-
Easy to integrate and customize. BrainFrame gives AI application developers and system integrators the following ways to develop and integrate: 1) You can build on top of it through the REST API, the Python API and the database API (BrainFrame API Documentation) and integrate it into enterprise or industrial applications. With the REST API, the time from a video frame arriving to the application being triggered can be a few dozen milliseconds; with the database API it is within a few hundred milliseconds. Database history can be kept permanently or purged periodically. 2) BrainFrame is based on the algorithm capsule specification. If needed, algorithm developers can use common development frameworks such as TensorFlow, OpenCV DNN or OpenVino to develop new algorithms, then easily package them into algorithm capsules themselves and load them into BrainFrame to enable entirely new vision applications. OpenVisionCapsules provides template programs and packaging tools. Packaging a new algorithm from a template usually takes fewer than 50 changed lines of code, and the capsule is generated automatically by the packaging tool. The open-source code and template examples are available from OpenCV's GitHub: GitHub - opencv/open_vision_capsules: A set of libraries for encapsulating smart vision algorithms.
-
BrainFrame's automatic algorithm fusion and scheduling/acceleration engine uses several advanced AI technologies and algorithms (patent pending): 1) Algorithm scheduling, resource scheduling and the AI inference pipeline are fully optimized to make full use of all system computing resources, with a complete user interface for control. Application engineers using BrainFrame no longer need to think about these issues and can focus on solving problems in their application domain. 2) BrainFrame is compatible with Intel CPU, Intel Graphics, Intel Movidius/FPGA* and NVidia GPU AI acceleration platforms, and can be extended to support or be compatible with other chip architectures. 3) The system fuses algorithms dynamically in real time using automatic algorithm fusion. Depending on the needs of the application, users can add the algorithm capsules they need at any time, or unload those they are not using for now, plug-and-play, without affecting the running system. If a specific new vision or neural network algorithm has to be developed for a special application, algorithm engineers can reuse the many released algorithm capsules and focus only on the specific algorithm they need. BrainFrame automatically fuses the new algorithm capsule with the other existing capsules dynamically in real time.
For system integrators, smart camera and hardware manufacturers, the powerful and easy-to-integrate BrainFrame platform makes it possible to enter the market quickly and to make the return on investment clearer and measurable, in the following ways: 1) Hardware manufacturers can build smart cameras according to the open-source algorithm capsule specification, and become fully compatible with BrainFrame through hardware algorithm capsule adaptation. 2) OEMs can also produce smart vision edge computers with BrainFrame preinstalled, for system integrators to deploy at scale in enterprise-level or city-level applications. 3) System integrators can use BrainFrame directly with already-deployed standard IP cameras. With simple configuration, intelligent video analytics operations can start quickly. They can also easily use BrainFrame + ordinary IP cameras + smart cameras to quickly deploy an edge/device/cloud hybrid computing architecture that uses the least bandwidth while making full use of edge, device and cloud computing resources, and build a complete system solution right away.
(* Note: This functionality will be released publicly in the future. If you are interested in early access, please contact our sales staff.)
BrainFrame and pre-installed algorithm capsules download (including documentation):
https://dilililabs.com/docs/downloads/
Open-source code for the algorithm capsules (BSD license):
Algorithm capsule documentation:
https://openvisioncapsules.readthedocs.io/en/latest/
OpenCV.org Hardware Partnership Program
https://opencv.org/opencv-partnership-program/
For cooperation and technical support, please sign in and ask here: https://forum.dililiLabs.com/
Email: info@dililiLabs.com
We believe AI gives people more opportunities and will make the world a better place. We are committed to innovating so that machines can see and understand the world. Our goal is to be the world leader in the rapid adoption of AI, serving customers in smart vision and intelligent video analytics.