Recently, extra concerns have been raised regarding child safety after the alleged abuse and neglect reporting incidents at a children’s caring home, demanding better monitoring and enforcement of the law. Meanwhile, the Omicron outbreak has also placed a heavy burden on public health care and isolation facilities, where a surge of infection cases has been recorded.
Thus, video monitoring appears to be an alternative, helping to ease the shortage of healthcare staff at both clinical and elderly facilities during the pandemic situation. Nevertheless, security and privacy remain some of the concerns of video monitoring that have to be resolved. In light of this, the Data Science Lab at the Department of Statistics and Actuarial Science (SAAS), The University of Hong Kong (HKU) has developed several real-time video analytics apps using the first Hong Kong-made edge computing AI chip donated by an HK-based technology company, making them able to detect human body movements (e.g., walking, falling, leaving the room, etc.) and facial expressions (e.g., crying, yelling, etc.).
Applying the techniques of bounding box detection, object detection, and motion classification, the team built the apps with “ResNet-32”, a Deep Learning Neural Network specified in image recognition, and “Flasks”, the framework of programming language Python, for video analysis, and used over 5,000 images on average collected from the Internet to train each model.
Densifying the data centre network, the edge computing AI chip allows faster computing, better data security, and efficient control over the continuous operation – it is ten times faster than the market edge-based AI chip in terms of computation power.
Since the video analysis are done in the AI chip itself, it does not require running the apps in a cloud computing platform, thus overcoming the security and privacy concerns of video transmission on the Internet. In addition, the AI chip can be implemented in any device such as robotic pet, surveillance camera, etc. It can be used in homecare/childcare/elderly care centres, offices, shopping malls, or hotels for risk detection and personal care monitoring.
In the case of child abuse at children’s caring homes, the video analytics apps can be deployed for childcare monitoring. Also, the apps can assist nurses in monitoring patient risks in isolation wards, alerting them if patients leave the ward, asking for help or deliberately self-extubate their endotracheal tubes, and helping elderly homes to detect elderly care risks such as falling and yelling for help, etc.
The CEO of the tech company that donated the chip is the first edge computing AI chip developed by the firm in Hong Kong. He thanked the Data Science Lab for deploying the newly invented AI chip with their real-time novel video analytics apps that overcome the security and privacy concerns on video analytics and surveillance.
Other than security and safety monitoring, the AI chip can also be used in the processing and analysing big data in finance and medicine, for investment and disease diagnosis decision-making, virtual reality, robot automation, smart home, physical and cognitive training, drug and gene discover.
The Deputy Director of the HKU SAAS Data Science Lab stated that the Lab is currently researching a novel parallel and distributed AI algorithm and is planning to deploy it in an environment of clustered edge computing AI chips. In this way, real-time big data analytics with supercomputing power can be done and is a breakthrough in AI development.
About HKU SAAS Data Science Lab
The project team of the Data Science Lab at HKU comprises Head of Department of Statistics and Actuarial Science (SAAS) Professor Guosheng YIN, Director of the SAAS Data Science Lab Dr Eddy LAM, Deputy Director Dr Adela LAU, as well as undergraduate and taught postgraduate students in the Department.