ComputerVision: Leveraging AI to Detect NJ Transit Vehicle Numbers

By Eric Huang and Abanoub Masoud

Scan the

QR Code

Click "Enable Camera" Then scan over this image

Click "Enable Camera" and click yes on all permissions

How does it work?

• Built on OpenVINO AI libraries
• Vehicle Detection and Optical Character Recognition (OCR) models
• By using lightweight open-source models, saves on development time and processing time
• Calls for information from NJ Transit databases
• 4 steps to go from image to info
1. Filter vehicles
2. Read text
3. Filter text
4. Display information

​

Try scanning this one too!!

STEP 1 – FILTER VEHICLES USING VEHICLE RECOGNITION MODEL

Selects the vehicle that is most likely to be a bus

Crops image to filter out noise like street signs, etc

STEP 2 – READ TEXT USING OPTICAL CHARACTER RECOGNITION (OCR) MODEL

Run OCR on cropped image

Create list of potential numbers

STEP 3 – FILTER TEXT USING RULES

Rules

• Numbers only
• Vehicle number is 4 or 5 digits
• Cannot start with 0
• …etc

Rules can be customized via regex

STEP 4 – CALLS API TO DISPLAY INFORMATION

YARD MANAGEMENT

REAL WORLD IMPLEMENTATION IDEAS

MONITORING AREAS

CONSUMER APP

• Currently, bus location in garage is tracked through a person with a clipboard
• By installing cameras, this could be automated

• Log where busses last were in busy areas such as the Port Authority Bus Terminal

• Allow consumers to see live tracking information, similarly to the MyTrain feature in the NJ Transit app

Thank you!

Any questions?