人工智能物联网研究院知识转移项目(只有英文版本)

Guangdong-Hong Kong-Macao Greater Bay Area Standards, T/GBAS 0003-2021, "Data for intelligent operation and maintenance of bridge-island- tunnel crossings-Bridge structures."

This standard specifies the data format for intelligent operation and maintenance of bridges in the GBA.

Guangdong-Hong Kong-Macao Greater Bay Area Standards, T/GBAS 0001-2021 "Data standard system for intelligent operation and maintenance of bridge-island-tunnel crossings-Guidelines on the establishment of standards."

This is the first-ever standard for the Guangdong Province GBA, organized by the HK-Zhuhai-Macao Bridge Authority. It specifies the data format for intelligent operation and maintenance of bridge-island-tunnel crossings.

PolyU Footbridge Health Monitoring System.

The system consists of 88 sensors of 13 type to monitor the environmental parameters, various loads, and responses of the bridge in real-time. A touchscreen mounted at the bridge end of Block Z and an Internet website provides an education platform for students and the public.

Low-cost Acoustic-based Liquid Fraud Detection supported by RGC Research Impact Fund "Tackling Grand Challenges in Food Safety: A Big Data and IoT Enabled Approach."

Acoustic techniques have been used for liquid detection but with expensive ultrasound devices. To achieve low-cost liquid fraud detection, we develop a low-cost method by measuring the acoustic absorption of the liquid using commodity acoustic devices. Our method can detect fake liquor, wine, and olive oil with an accuracy above 92%.

We are adopting the research outcomes for the application of a project with The Hong Kong Sports Institute.

Low-cost Hyperspectral Sensing for Food Freshness Detection supported by RGC Research Impact Fund "Tackling Grand Challenges in Food Safety: A Big Data and IoT Enabled Approach."

RGB imaging can be used to detect food quality but is not accurate as it can not capture minor and subtle details of food images. Hyperspectral imaging (HSI) systems can capture the spectrum of each pixel of a food image but are highly expensive and available only in laboratories and industries. We develop the first smartphone-based low-cost hyperspectral imaging technique without embedding any external sensor to achieve low-cost hyperspectral sensing for food freshness detection. Our system can detect food freshness with 90% - 92% accuracy.

Autonomous Cooperative Multi-robot System: A Fully Distributed Approach

We aim to develop intelligent robots that can learn cooperative strategies from interactions with others. Specifically, we investigate the challenges of learning cooperation in a distributed way, such as: partner modeling, partial observation, and large-scale multi-robot system.

We are adopting the research outcomes for: 1) application of a departmental demo funding project; and 2) application of an STF (smart traffic fund) project.

Distributed Edge Intelligence for AI-empowered Applications

A significant problem in real-time video surveillance is providing advanced analytical services while reducing latency. We develop a distributed edge intelligence system with multiple edge devices (i.e., Jetson Tx2, Xavier NX) sharing both data and computation resources to support collaborative real-time video surveillance. We deploy lightweight deep learning models on resource-constraint edge devices to provide pedestrian tracking and reidentification services. To reduce the latency, we designed a scheduler running independently on each edge device to enable sharing of the computation resources. The scheduler dynamically offloads the computation tasks within the cluster whenever the device is overloaded. We build a real-world prototype deployed in an indoor environment, showing near-real-time performance. Our system has many benefits, including low latency, low cost, higher scalability, and increased data privacy.

We are adopting the research outcomes for: 1) application of a departmental demo funding; 2) application of an MHKJFS (Mainland-Hong Kong Joint Funding Scheme) project; 3) collaboration with PolyU CFSO for campus video surveillance; and 4) collaboration with Queen Elizabeth Hospital for hand hygiene monitoring in clinical environments.