intelligent computer vision agents optimising pti safety ... · computer vision based approach...
TRANSCRIPT
Intelligent Computer Vision agents Optimising PTI Safety and Train Dwell Times Lancaster University
Dr Gruffydd Morris, Professor Plamen Angelov, Lancaster University Dr Howard Parkinson, Digital Rail
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Project Scope
Challenge: To assess the feasibility of Computer Vision (CV) improving safety standards at the PTI while reducing impact of passengers on dwell times
3 Work Packages:
1. Asses the scope of the current issues surrounding PTI safety and impact of passenger boarding / alighting on dwell times
2. Through mode of prototype, test the capability of CV to augment safety at the PTI
3. Again through prototype, provide an analysis of busyness on trains and platforms with the goal of impacting dwell times
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Project Scope
Solution to augment information available to users and stake holders
Detect object and users in or approaching the PTI.
Detect and identify users and objects on-board carriages.
Computer Vision based approach using surveillance systems
Detect and identify individual objects
Intelligent source
integration
Key: - Platform based - Carriage based
Station Staff Information
Passenger Information
Train Staff Information
Detect and identify individual objects
Passenger Information
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Findings: Stakeholders have different requirements
Driver:
– High risk / on track info on approach
– Passengers near doors on dwell
– High risk (red) on depart
Platform Staff:
– Busyness on approach and dwell
– Simple alerts – what is happening where on the platform
– Safety info on depart
Control room Staff:
– Busyness only when too high
– Safety screens
Passengers
– Only busyness on approach and dwell. Nothing on depart
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Findings
Can detect where passengers are on the platform (if they cross the yellow line)
Alerts can be sent to drivers, staff and other stakeholders if someone is stuck in the gap
Can get a measure of the number of passengers on both the train carriages and the platforms
All using existing camera infrastructure
Processing can be done in real time on a computer the size of a credit card (Raspberry Pi for example)
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Limitations
Shadows and sun glare can interfere with detections
For gap detection, a yellow line (or similar) lengthways down the image is required
Camera perspective can be an issue if fish lens or similar is employed (regular key stoning can be compensated for)
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
How it works: Gap Safety
Uses CV analysis with filters applied to remove clutter
Computationally efficient
Detects static and moving objects
Distinguish between objects based on motion criteria
An example of the detection criteria of people and motion, and distinguishing between zones.
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Busyness information
Both show platform loading information
–Top image: quiet around the train and quiet on the platform
–Bottom image: Shows busyness increase on platform and slight increase near train
This information set does not include gap analysis
Busyness behind the line
Busyness behind the line
Busyness over the line
Busyness over the line
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Under the hood: Gap analysis combined with busyness
Busyness behind the line
Busyness around the line
Busyness over the line
Time over the line
Busyness around the line
Busyness over the line
Busyness behind the line
Time over the line
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Under the hood: Busyness on Platform and Carriages
Analyses both platforms and carriages
Provides a busyness measure for each, compensating for perspective distortion
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
What a Stakeholder might see: Drivers
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
What a Stakeholder might see: Staff
ALERT : PLATFORM 1 ZONE 2
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
What a Stakeholder might see: Passengers
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Stakeholder benefit
Passengers (customers)
– improved safety at the PTI
– optimisation of their PTI experience
Drivers
– increase drivers experience and go some way to reduce stress and pressure.
– assist the driver to make accurate decisions
– assist the driver to maintain on-time services by reducing passenger boarding dwell time, further helping to reduce safety related incidents.
Train and Station operators –
– Improved safety of passengers through a reduction in PTI FWI incidents
– faster passenger boarding at the PTI can aid station operators in running a safer and quicker service.
– (side effect) improved passenger movement will help to reduce larger crowds and subsequently improving safety
COF-PTI-04 Intelligent computer vision agents optimising PTI safety and train dwell times March 2018
Suggested future work
Stakeholder engagement to improve design and fit in with current workflows of the industry
Pilot programme would be useful to understand how stakeholders interact with the system
Human factors study, in collaboration with work done on other projects such as CONSIST
Questions and Thank you