The train for the Automated People Mover (APM) in Hong Kong International Airport, it is taking passengers amongst Terminal 1, Midfield Concourse and Sky Pier, will normally be operated in an Automatic mode (i.e. driverless). But, for the train under maintenance at night, a train operator will drive the APM in a Manual mode.

In order to prevent incidents made by the train operator during the manual drive, this project aims to develop an AI system with video analyzing for prevention of incidents arising from Signals Passed at Danger (SPAD) and other human factors by the driver. In this trial, we will setup four sets of device in APM trains under this project.

The device shall be portable and comprise of key components including the number of cameras, buzzer, siren, SMS module, internal battery & data storage for video recording (for incident investigation) both lasting for 6 hours, and the device will be easily installed or removed day by day in the train cabin with the key features to detect dangerous behaviors and abnormal obstacles in the railway.

Project Challenges:

1. Being able to detect unsafe / dangerous driving behaviours or environment of the video captured by cameras (integrated-standalone unit).
2. Being able to generate the corresponding alarms (audio and visual) for different unsafe detections.

• The physical device shall be portable, easy to carry with around 2kg in weight and around 70mm (Length) x 70mm (Depth) x 240mm (Height) in size of one integrated-standalone unit.
• The device with AI video analytics will detect the environment of APM tunnel, signal lights and the dangerous behaviors of the train operator. For different detections, it will trigger corresponding alarms, sound and light to alert the train operator and notify personal in the station control room via SMS message.
• After all detection and alarms for each use by the train operator, all these information/logs stored in the device including the video records will be uploaded to the off-site computer for video analyzing against driver response(s) & images together so as to facilitate the incident investigation.
• The solution shall include a 2-year warranty after the implementation for both hardware, software and maintenance services on site
• The physical weight and size in one integrated-standalone unit and the AI for video analyzing are key successful factors for this solution.