27 Jul HK’s AI crowd management for more efficient services
A Hong Kong start-up is reportedly helping universities and railway operators in Hong Kong better manage their crowds as private businesses and the government in the former British colony step up their use of artificial intelligence.
The City University of Hong Kong is deploying the company’s EdgeAI system, which runs algorithms on an industrial computer to process images of people and objects, to count the number of students and teachers on campus to provide insights into crowd flow at major entrances and exits, according to the firm’s Founder and Managing Director.
The technology is used in the city’s subway stations to provide an estimate of passenger waiting times during peak hours, while local gas companies deploy the AI system to detect entries to remote sites as well as leaks or rust on gas pipes that need repair.
The start-up was founded in 2009 as a software development firm for public organizations and international brands, but its head noticed saw the increasing potential in the application of AI. Thus, the AI branch was founded two years ago to focus on using perception-based AI technology to help businesses detect people and various objects.
The relevance of similar start-ups is a result of Hong Kong’s increasing adoption of AI in businesses and public organizations. This is because the government is working to encourage innovation and build a world-class smart city in mobility, living, environment, public service, and other fields.
The government saw the usefulness of the start-up’s crowd-counting technology after the 1 July protest march in Hong Kong. Teaming up with a social sciences professor at Hong Kong University, and a geography researcher from Texas State University, the start-up set up three iPad cameras on Percival Street in Causeway Bay and four more on Arsenal Street in Wan Chai.
Using the technique of object identification, the system tracks protesters when they appear within the camera frame and counts them as they cross a counting line. One camera covered three lanes and Wong adjusted the locations of the other cameras to avoid overlaps. The algorithm program ran directly on the iPads rather on the company’s industrial computer.
The project concluded that 265,000 people joined the protest on 1 July, while organizers said 550,000 people marched and the police estimated it was 190,000.
The takeaway is that AI is more objective and accurate than manual counting. The numbers generated merely serve as a reference to the public. It is hoped that the AI can be trained to run better on smartphones and enable accurate counting through this project.
Another AI counting project is being developed for when the Civil Human Rights Front plans another protest march from Causeway Bay to Admiralty.
Since June 2019, Hong Kong has seen a series of record-breaking protests as locals marched against an extradition bill proposed by the Hong Kong government that would have allowed the transfer of suspects to jurisdictions with which Hong Kong has no extradition agreement, including mainland China.
Earlier this month, Hong Kong’s Chief Executive declared the highly unpopular bill to be “dead.” However, protesters continue to demand a total withdrawal of the bill and for the CE to step down.
The city’s tech-savvy protesters have taken extra precautions to hide their identities during marches by ditching personalized metro cards, switching to one of the most encrypted messaging apps available, wearing masks and goggles, and refraining from posting selfies on social media in case they can be identified by law enforcement.
It was noted that the aforementioned AI project carefully handles privacy issues as it blurs any faces that are captured and deletes all data afterwards.
Project managers have stated that they have been extremely careful in regards to the technology deployed. No facial recognition tech was used and, in fact, the tech deliberately shoots images from the back and does not share any of the videos with the third-party volunteers. The objective is to merely count for a number.