Singapore organisations show how analytics, AI and the IoT will impact our lives in the near future

Earlier this month, Singtel, Nanyang Technological University (NTU) Singapore and the Agency for Science, Technology and Research announced that they will collaborate to accelerate innovation in the fields of artificial intelligence (AI), robotics, advanced data analytics and smart computing to create solutions that will help local and global enterprises in their digital transformation journeys.

A number of applications for public safety, transportation, healthcare and manufacturing industries were showcased at the launch event:

Smart sensing

IntelliSURF by Singtel
Featuring advanced video analytics, 3D visualisation, complex events processing

Singtel’s IntelliSURF is an agile, modular smart city platform that enhances situational awareness through an integrated dashboard for greater command and control. It allows data from various sensors to be ingested, analysed and visualised in 2D/3D forms on a single dashboard. The platform provides an immersive experience and automates processes, workflow and reporting to help organisations respond faster and perform operations more efficiently.

CLARITAS by Singtel
Featuring AI (facial recognition), intelligent edge sensors

CLARITAS is a smart surveillance camera with edge analytics embedded to automatically process real-time video images at the camera end. This reduces the burden of sending high volumes of data over the communication network to the backend system. Through speedy facial recognition at the camera’s edge, CLARITAS is useful for track-and-trace of people-of-interest in secured areas and for managing visitor queues.

Cross Camera Human ReIdentification by NTU
Featuring AI for facial recognition, visual feature identification and deep neural networking

Automatically identifying persons of interest from camera surveillance.

A computer, unlike humans, cannot easily recognise the same person caught at multiple camera angles in various locations. NTU researchers have developed AI software that enables a computer to identify the same person from various angles. This NTU algorithm could help investigators improve productivity and save time, as it minimises the manual labour of reviewing hours of video footage.

For instance, given the back view or side profile of a person, the software can scan prerecorded footage on other cameras to identify the same person. To develop this technology, the team fed the AI with 12,000 images as training data, successfully training it to distinguish between 750 different individuals.

Visual Object Search by NTU
Featuring AI (visual feature identification)

Using a logo on clothing to verify that the same person is in separate camera footage.

Searching for a person can be improved with a logo search function. If a person-of-interest was wearing a tshirt with an obvious logo, databases can be searched for that person using an image of the logo. This can narrow down a list of suspects, saving time for investigators.

3D FaceCollage by NTU
Featuring AI (facial recognition), 3D visualisation

FaceCollage can fill in the blanks for a partial view of a face using standard off-the-shelf hardware.

FaceCollage can generate an instant 3D model of a person’s face through an affordable consumer-grade depth camera like the X-Box Kinect. In the past, such results would only be possible with expensive equipment and only after training the AI for a long time. This user-friendly system is easy to deploy and does not need any complicated calibration. The 3D face model can be used to generate more visual angles to a subject’s face in different poses and expressions, which can be analysed for use in security or healthcare applications. For instance, a 3D-generated photo can be used to search for a subject’s face in a database of videos or photos. 

Audio Intelligence Monitoring at the Edge (AI-ME) by NTU
Featuring acoustic signal processing, audio intelligence, deep learning

Explaining how AI-ME works. The system can detect abnormal sounds such as gunshots.

The graph on the top left accurately reflects whether the ambient sound is speech or something else.

The AI-ME edge device.

Low-cost intelligent sound sensors could soon help to monitor the environment, picking up calls for help or to monitor road safety in conjunction with video cameras. Consisting of a microphone, battery pack and a processor equipped with an AI algorithm, this sensor can detect a variety of sounds and categorise them from speech and coughing to claps and water being poured.

One potential use for these sensors is to identify sounds of an accident or a car crash and alert the relevant authorities. Another could be to enhance the smart home experience, for example monitoring the elderly without being intrusive. It could be useful if someone falls and shouts for help. It can also be used to gain insights into their state of health, such as how often they are coughing and if they are drinking enough water.

Edge intelligence refers to processing data at the point of data capture, instead of sending it back to a server to be analysed.

Ageing-in-Place with Unobtrusive Sensing and Data-driven Storytelling by NTU
Featuring AI, the Internet of Things (IoT), intelligent edge sensors, data analytics, machine learning

Sensors can track an elderly person in his or her home. Watch the Instavideo of the ageing in place demo.

The storytelling dashboard rates an elderly person's health and wellbeing.

By outfitting a home with a variety of sensors, such as motion-activated sensors, infrared, light, humidity, pressure or switch sensors, NTU scientists are looking to gain more insights into the life of an elderly person and to improve their wellbeing. Through predictive data analytics and the IoT, scientists may be able to detect previously-hidden problems faced by the elderly that have escaped their caregivers.

For instance, if an elderly person sleeps longer than usual or sits on the sofa for an extended period of time, it could be an indicator of loneliness or an illness. Likewise, if he or she opens and closes doors or drawers repeatedly, it could indicate anxiousness or forgetfulness.

All the analysed data can be put together in a digital dashboard with an automated storyteller, which could paint a comprehensive picture of an elderly person’s health and mental wellbeing.

Smart productivity

Isera - Mobile Service Robot with Artificial Intelligence by the Institute for Infocomm Research (I²R), A*STAR
Featuring robotics, location-based navigation, intelligent sensing, natural language processing, machine learning

The Isera robot. Watch the
Instavideo introducing Isera.

Isera is an autonomous service robot that can understand speech commands. It can also interact with humans and assist the person on the request. It can navigate autonomously when given a target destination. Isera can be used in the healthcare and hospitality industry to attend to service request from in-patients or customers respectively. It can also perform the duties of a night guard to conduct security monitoring around the controlled premises.

Cobot (Coworker Robot) with Artificial Intelligence by I²R, A*STAR
Featuring AI, robotics, haptic learning, machine learning, speech recognition, natural language processing

A cobot is a robot designed to work together with humans. This robotic platform can be taught to perform a task by demonstrating it and giving verbal commands. The robot undergoes haptic learning, which relates to the sense of touch, in particular to the perception and manipulation of objects using touch and proprioception*.

AI in the robot allows users to verbally instruct the robot and make adjustments in a manufacturing project without the need for an expert to change code or otherwise reprogram the robot. The robot is able to learn the new process using a shadowing function**. This solution supports rapid product development and improves manufacturing productivity.

This robot can learn by watching others.

Software Bot by Singtel
Featuring software bots, machine learning, process automation

Before using the Software Bot - more steps in the process, and order processing took more than 18 days.

After using the Software Bot, there are fewer steps in the process and order processing now takes under five days. Full-time equivalent*** costs have fallen by more than half, allowing Singtel to bring this work back onshore. 

A software bot uses robotic process automation (RPA) to help automate some of the mundane tasks that humans do, increasing speed and accuracy. Any organisation which wants to automate administrative processes, repetitive and voluminous tasks can make use of a software bot to improve operational efficiency. 

Real-time Digital Twinning of Workplace by Advanced Remanufacturing and Technology Centre (ARTC), A*STAR
Featuring robotics, 3D virtualisation

The digital twin on the screen shows readouts from various sensors, and mimics the physical robot's movements exactly. The data can be used for CMBx.

This platform monitors physical machines, in real-time, using virtualisation mechanisms. It emulates the physical robot with a digital twin on screen and streams robot movement onto the screen in real time. This allows the behaviour of the robot to be monitored without direct physical presence or interference with the manufacturing process that is taking place. Events can be played back for root cause analysis.

Digital twinning allows safe, unconstrained observation of machines (assets) in heavy-duty manufacturing environments, and helps manufacturers increase productivity. Exercising control over the physical asset via the digital twin is also a future possibility.

Condition-Based Maintenance by Singtel
Featuring AI, IoT, intelligent edge sensors, predictive analytics, machine learning

Explaining condition-based maintenance using a fictional escalator. Watch the Instavideo showing what happens when the escalator looks like it will fail because of an overheating sensor.

Condition-based maintenance (CBMx) relies on a combination of technologies such as IoT sensors, advanced data analytics and AI to predict and prevent failure of equipment. It involves the use of predictive analytics and machine learning to determine the frequency and factors of failure in expensive customised equipment, thereby allowing effective preventive maintenance to be performed before a failure incident actually occurs. Any industry that employs a lot of equipment for daily use can benefit from condition-based maintenance. The more complex the equipment, the larger the resulting savings.

Real-time multilingual speech recognition by NTU
Featuring advanced speech recognition, natural language processing, machine learning

Code-switching refers to switching between multiple languages within the same sentence.

Actual speech converted to text in real-time. Watch the Instavideo of speech recognition in action.

Current digital assistants, such as Apple’s Siri, Microsoft’s Cortana and Google Assistant may find it hard to understand the speech of a Singaporean, as multiple languages may appear in the same sentence whereas most speech recognition systems only process monolingual speech.

A new algorithm was designed by NTU scientists to recognise conversational English and Mandarin phrases in the same sentence. Using AI and machine learning techniques, the software can convert this mixed speech to text in real time.

This state-of-the-art system will be useful in applications such as chatbots used by companies to interact with their customers, transcripts of call conversations and also automated captions for videos.

Explore:

Read the TechTrade Asia announcement of the collaboration

*Proprioception refers to being able to tell where the body is in a space and the ability to avoid colliding with objects in that space.

**In shadowing, a robot can copy motions from a human or mechanical teacher and record the motion sequence.

***FTE is an SAP measurement of headcount that takes into account the number of hours an employee is expected to work among other things.