興倫人工智能有限公司Hanlun Artifical Intelligence Limited

Noise leakage

The Noise Leakage Visualization project is a commercial attempt to build with industrial packages to better utilize sophisticated measurement devices. The post-processing capabilities result in magnitudes of enhancement in signal resolution and the output can be projected to visualize sound leakage.

Noise leakage

The visualization concerns more on Spatial Domain than Temporal Domain in regards of continuity. (Click here for more details)

Noisense

The Continuous Monitoring System for Noise Sensitive Receivers near construction sites (codename: Noisense) automatically streams audio to be processed for sound pressure level computation and fast fourier transformation before being passed to a neural network for noise source identification. Results can be visualized on a web dashboard.

Noisense

Features

• Integrated with type I microphones
• Instant messenger notification
• Abnormalities detection
• Video streaming as alternative data (demo)
• Hybrid servers to distribute computation expense

Expertise

• Devices: CentOS
• Streaming services: FFmpeg
• Motion detection: OpenCV
• Web services: MySQL, Node.js, Socket.io
• Machine learning: PyTorch

Noisense

End-to-end data pipeline

1. Audio collected by class 1 microphones was streamed to cloud servers
2. A-weighted sound pressure level was calculated with Python
3. The audio was analyzed and classfied with a neural network
4. Calculated level and audio class results were uploaded to a database
5. Web server fetched the results and displayed them on the web dashboard

Noisense

Deep convolutional neural network

1. Employees recorded the time and labelled the noise source on site to prepare for neural network training
2. Labelled audio was transformed to spectrograms
3. ResNet was retrained with the labelled spectrograms to achieve ability to auto classify

HZL

The Continuous Monitoring System for large water bodies will be a collaboration between universities and us involving both Hong Kong and Mainland parties. Please stay tuned!

MOOC Platform

Starting from the Open edX MOOCs platform under the Python-Django framework, we developed a cloud-based management platform with both a learning management system with real-time transmission of learning content, generation of assessments and automated grading; and a content management system with an editing studio. We also integrated the system to a data visualization dashboard for dynamical student performance feedback.

MOOC Platform

Features

• Easy-to-Scale learning and assessment platform
• Control of student flow with prerequisite knowledge determination
• Built-in studio for agile course development
• Numerous sensors on student footprints for analysis
• Suitable for carrying out educational experiments

Expertise

• Framework: Open edX, Django
• Database: MySQL, MongoDB
• Data Visualization: Power BI, D3.js
• Analysis: TDA for time series data
• Recommendation Engine: Scrapy, Jieba, TensorFlow, DKN

MOOC Platform

Quantitative data analysis

• This is part of the DevOps cycle that the collected data will be basis of algorithms to emerge intelligence for platform upgrade
• Accompanying data collection includes learning footprint with timestamps and some tailor made KPIs
• Tailor-made KPIs are introduced for student performance analysis; sophisticated methodology is employed for time series data:-
• Regression and survival analysis are employed for education experiments

MOOC Platform

Knowledge graph visualization

• Visualization for Client's acyclic directed knowledge graph
• Interactive dashboard for prerequisites information

MOOC Platform

Recommendation engine

• Data from world-wide sites were scraped for natural langiage processing
• Processed with Chinese language segmentation and keywords extraction
• Graph attention neural networks for recommendation
• Recommend learning modules by learning behaviour and predetermined prerequisites structures

Curriculum Generations

To share our vision to a broader audience, we actively participate the development of curriculum towards the next generation. Therefore we've started writing an ICT curriculum to equip students with the latest knowledge and skills of ever-advancing technology.

Curriculum Generations

Curriculum

• Artificial intelligence
• Object-oriented programming in Python
• Computational thinking
• Network technology
• Computer graphics generation
• Etc.

Curriculum Generations

Features

• Interactive programmable environment: Colab
• Nurture of web searching skills: Redirecting to search engines

Case studies

• Donkey Car (optical flow, heuristic function)
• Sudoku (computer vision, constraint satisfaction)
• Etc.

Curriculum Generations

Some other generations of curriculum can be found in our github page

Next Gen Curriculum

Machinery DevOps

We've been serving in the domain of sports science since the beginning. Over the years, several DevOps projects were completed to facilitate the usage and efficacy of Tai Chi exercise machineries, which act as a partner of 'Push-Hands' while providing real-time feedback to the practitioner.

Machinery DevOps

Curve detection & processing toolbox

Our Client owns a mechanical motion recorder and different versions of motion player to provide practitioners a guidance point’s trajectory. Important DevOps involves include data denoising; curve smoothing; motion curve visualization; troubleshooting; and data transformation for the compatibility, convention of coordinatizations and data formats for the interuse of these machines.

Machinery DevOps

Monocular motion sensing & analytics system

While our Client is a mechanical motion player giving guidance by constrianted on trajectory of a hand, Tai Chi emphasizes on motion of the whole body; therefore this system was developed to provide whole body motion monitoring and analysis by a single webcam.
Later, our R&D project concluded that visualization of the center of gravity could serve as biofeedback to help practitioners and hence was desginated as one focus of the system.

Machinery DevOps

Plantar pressure sensing system

To further study the external force of a practitioner, we developed a software system that could boost the performance of the hardware designed by our Client using techniques transferred from the field of computer vision.
More specifically, we borrowed the idea of super-resolution to enhance the visualization of the plantar pressure from pixelated input to vector output. We also computed the center of pressure to serve as biofeedback for practitioners.

Machinery DevOps

DevOps for ergonomics

In an upgraded version of the mechanical motion player, ergonomics was the focus. To help our Client to make the mechanical motion player be responsive to a practitioner's height, we worked in two aspects: a height-measuring mechanism by computer vision to be input to the motion player, and an adjustment of the machine motion trajectory in scale for the player's output.

Machinery DevOps

Expertise

• Framework: Eel, HTML, Javascript, Python
• Database: SQLite
• CSS: Bulma, Bootstrap
• Visualization: Canvas, D3.js
• Machine Learning: Tensorflow.js, PoseNet
• Communication: TCP/UDP, WebSockets programming

Machinery DevOps

On going/ Coming soon...

• Temporal data analysis on derived quantity ("Qi") from time series