The department is looking for live captioning technology that can convert speech into text and display on the screen for audience reference. With real-time translation service, audience with different language preference can read the caption in their preferred languages for better communication without human interpreter arranged on site.

Information to be collected for each item include description of the commodity/service item, name/type of retail outlet, date of purchase, amount spent, delivery charge (if can be split), etc. In order to enhance the operation efficiency, respondents may choose to provide the hard copy receipts or upload photos of the receipts through the website instead of filling in the relevant parts of the questionnaires. It is time-consuming for coders to read the information on the receipts and input them into the system. In order to save manpower, it is proposed to extract required information from the receipts and output the information into a systematic format through AI. In future, we may also expand our usage to other types of documents. Therefore, it will be preferred if the proposed solution also has potential to extract information from other types of documents.

Our hotline receives a large number of public enquiries, especially during peak seasons. Therefore, we are looking for an Automatic Speech Recognition solution to reduce manpower and time on listening to voice messages and understanding the content as well as recognising key words and assigning the enquiries to appropriate officer based on the keywords recognised for follow-up. Officers could speed up preparation of replies, based on the automatically transcribed texts of the enquiries from voice messages.

Customised hardware product is required to tackle high temperature and humidity inside the compartment of the narrow and tight environment of a smart lamppost.

The narrow and tight environment inside the smart lamppost compartment is difficult to fit more devices of different functions together. An all-in-one processing computer is required to equip with a LoRaWAN end node module, UPS module and RS485 ports for LoRaWAN network connection, emergency power supply and connection to multiple sensors. A software module is also required for sending data through the built-in end node.

A hardware product is required to collect data including temperature, relative humidity, tilted angle, electric current and door open-and-close state for monitoring the environmental healthiness of compartments inside the smart lampposts

As a matter of background, our electronic Grant Management System holds a large number of restricted documents submitted by external users. We would like to seek solutions on full text searching functions in multiple files, which should be applicable to the following: Encrypted PDF file in the database Encoded PDF file in the database (binary-to-text encoding) Encoded Microsoft Word file in the database Text searching solutions which support Optical Character Recognition (OCR) Text searching in the database's data field All encrypted PDF files belong to “Restricted” in data classification.

There are Government water sports centres that offer a variety of water recreational activities, providing the public with a diverse range of options to enjoy the coastal environment. There have been reports of serious accidents like vessel collisions and stranding, and potential dangers of drowning, which could lead to health hazards such as hypothermia and heart stroke, and requires immediate rescue and medical attention. Other possible risks may include adverse weather, equipment failure, reckless behaviour or lack of experience. These reports have highlighted the need for putting measures and supervision in place to ensure participants’ safety in water activities. To assist the water sport staff and ensure the safety of participants, AI-powered Long-Range Video Analytic camera can be a useful tool for monitoring and managing water areas, providing an enjoyable experience for all, with the followings proposed objectives to achieve: - To propose and setup one or more outdoor AI-based Long-range Video analytic camera to cover the designated activity area. To detect potential hazards such as approaching storms, dangerous currents or submerged objects; and generating alerts to lifeguards or venue owners to take necessary precautions by AI and Open Data sources. To identify any erratic movement or individuals who have strayed outside the designated activity area, and sending notifications to operators for immediate action to prevent accidents or injuries. To count the number of kayaks, sailing dinghies and windsurfing boards users entering and leaving the designated area, monitor users on the water, and capture their return to ensure the safety and accountability of all users. To generate reports to venue managers for program planning and people management.

Signboard control in existing buildings is crucial for public safety especially for pedestrians and building occupants. It is also important for city management. However, the modus operandi of prevailing signboard survey (i.e., visual inspection, paper desktop study, and manual record-keeping) is time-consuming, labour-intensive and generally inefficient. To improve the efficiency and working environment of signboard survey, innovative technology with vehicle equipped with Artificial Intelligence (AI) may help to develop a real time visualised result for BD to make decision by professionals and further development of a smart signboard survey system. The AI will be of great assistance for a quick view of the streets-side for the potentially risky signboards. The AI designed will collect visual data via cameras into an edge-device and further identifying risky signboards using AI models. The first challenge for the AI, as expected to be proposed by technology ventures via existing 6-month PoC development, will be separation of signboards (together with its frames and support anchors) from the noisy street background. With this AI-supported data engineering, further analytics and observation will be made easier and more effective. Hence, as a milestone for further development, this project is scoped to support the AI-powered data engineering process by extracting signboards and their supporting frameworks. By completion, it would serve as a great launchpad for further development.

The project is divided into two phases:- To develop AI tool that transforms metadata from BIM, BIM-AM and integrated Building Management System (iBMS) into semantic knowledge graph. Data from another building will be used to validate the model; To demonstrate use case to optimise chiller plant efficiency using semantic AI approach (i.e. In other words, an AI algorithm will be developed to enhance chiller plant efficiency without E&M knowledge). We propose to use metadata from BIM and integrated Building Management System (iBMS) as the input parameter. An AI algorithm will be developed to calculate the correlation between the input and energy output and build a digital twin to simulate the situation of the chiller plant. Not only can it verify the efficiency of the existing library, but it can also facilitate the development of digital twins in EMSD.

This project aims to develop an artificial intelligence (AI) technology, known as LiDAR-based fiducial marker positioning system, to provide accurate positioning in GPS-denied tunnel environment for tasks such as tunnel inspection in public works projects. The objective is to address the critical business needs for safer and more efficient tunnel inspection in the confined and hectic tunnel construction environment. In the Trunk Road T2 and Cha Kwo Ling Tunnel project (the T2 project), “3S Tunnel Defect Inspector” – an AI drone system with onboard AI-based edge computing devices, which is the first AI drone system in Hong Kong for tunnel defect inspection, was deployed for efficient and automatic real-time defect inspection of TBM tunnel lining segments, viz concrete crack, concrete spalling and water leakage during the construction stage. This is a breakthrough to the conventional visual inspection in minimising safety risks of working at height, reducing tedious and repetitive tasks for human and enhancing tunnel construction logistics. However, the “3S Tunnel Defect Inspector” requires a unique QR code on each tunnel segment and an established tunnel BIM model to determine the location of observed defects. The application of this AI inspection system is suitable for use in most new tunnel projects for which the two prerequisites would have been allowed for in the contract. However, this is limiting the broader application of the system. Our project here is to bring a generalised solution for real-time positioning in tunnels for mobile robots that are assigned to carry out position-related tasks, by leveraging AI identification of fiducial marker, LiDAR 3-D point cloud models and an AI platform for evaluating and correcting the drift associated with the SLAM (Simultaneous Localization and Mapping) technique, such that the mobile robot can work out its own correcting maneuver to improve its position accuracy. This real-time positioning method can be applied to determine both the locations of inspection robot and the defects observed in tunnel, with the benefits of enhanced safety, more efficient inspection process, minimal installation required, high adaptability to all tunnels and low maintenance cost. (Note : If necessary, LiDAR scanning and construction of 3D Point Cloud Model can be carried out by the Trunk Road T2’s on site survey team to facilitate the PoC development. Equipment employed will be (1) Leica RTC360 (tripod-mounted laser scanner) and/or Leica BLK2GO (handheld laser scanner), using the supplied software to construct the 3D point cloud model.)

It is the principle duty of the Marine Police to maintain law and order in the HKSAR waters and protect people’s life and property. As one of the maritime rescue units of Hong Kong, Marine Police plays a key role in performing maritime search and rescue operations (SAR) that provides 24/7 emergency response capability to locate, assist and evacuate persons in distress, regardless of weather and sea conditions. The proposed system is to enhance the SAR capability utilising the existing land-based or ship mount EOS / camera system with an AI-driven search function that can search, detect and identify a person / vessel in distress at sea. The system comprises a network of high-resolution land based cameras and mobile cameras mounted on ships or drones, along with an AI-based software that can control and coordinate the networked cameras to perform sea surface sweeping and analyse the captured images in real-time. The main objectives of the proposed system are: To detect and identify a person / vessel in distress at sea; To enable the network cameras to sweep the sea surface in a coordinated manner according to the optimised search route formulated by an AI algorithm; To enable multiple cameras to collaboratively detect the person / vessel in distress by analysing the same scene from multiple angles; To alert the rescue team on the location of the person / vessel in distress; To provide real-time updates on the person's location and situation to the rescue team. The AI-powered automated camera system offers several advantages, including conducting search operations 24/7 with minimum human involvement, increasing resilience to adverse weather and sea conditions, improving efficiency of existing camera systems, reducing manpower demand and elimination of human fatigue. Furthermore, such a system can be easily expanded to include additional search capabilities (e.g. illegal immigrants and smuggling) in the future.