AI-based Traffic Enforcement
A compact, cost-effective, and scalable edge device to automatically capture traffic violations (including in cycle lanes).
As cities build dedicated cycle lanes to keep cyclists safe, most face the challenge of other road users not respecting or illegally occupying these lanes. This tends to cause conflict between cyclists & motorists, defeating the whole purpose of creating safe infrastructure for cyclists.
This is especially a challenge in Indian cities, where law enforcement is already overburdened and unable to effectively enforce traffic laws & regulations. Hence, in January 2022, the State of Karnataka invited startups to develop contactless cycle lane enforcement solutions that would allow cities across the state (and country) to ensure cycle lanes are used safely & effectively.
After a competitive selection process, the solution proposed by us (through Urban Flow) was awarded a grant to design, pilot and manufacture an AI-based cycle lane enforcement device. This project is currently in the pilot stage.
How it Works
We've built a truly unique device that's tailored to the needs of the Indian context where cost-effectiveness & scalability are crucial. With an edge-computing based architecture and 4G LTE connectivity, it's designed to be compact, easy to deploy, and requires minimal effort for setup and installation.
Once installed, our device performs the following steps for cycle lane enforcement:
-
High-resolution camera captures video feed of cycle lane
-
AI-algorithm detects & classifies vehicles, flagging those that are illegally operating in the cycle lane
-
AI-algorithm identifies license plate of motor vehicle and performs ANPR (automatic number plate recognition)
-
Evidence image, along with the vehicle's license plate number and other details about the violation are uploaded to our servers
-
Violation details are transferred to the traffic police authority for generating a ticket/penalty and further processing
Detecting Violations
Leveraging our past experience of building AI-based devices such as the Live Bicycle Counter, we started by creating a large dataset of images to train state-of-the-art (SOTA) AI based object detection models.
With robust data from a wide variety of street conditions and cycle lanes, we built a model that is capable of accurately detecting & classifying upto six types of vehicles.
-
Bicycles
-
Motorcycles
-
Autorickshaws
-
Cars
-
Buses
-
Trucks
The video stream is processed in real-time through our violation detection model and if a motor vehicle is identified to be violating the cycle lane, an evidence image is captured.
Automatic Number Plate Recognition (ANPR)
For automated traffic law enforcement systems, it's crucial that vehicle license plates can be read in a fast, accurate and reliable manner. Without this, the advantages of such a system are limited as manual intervention is required for identifying every violator.
In the Indian context, building an ANPR system has been quite the challenging process. We've experimented with various existing OCR libraries & frameworks, building our own image processing pipelines, and using pre-trained AI models. However, with many Indian vehicles having non-standardised license plate sizes, fonts, and shapes, none of these existing methods worked with any degree of reliability.
We then created a large custom dataset which would be robust enough for training our own AI models. Our model now accurately reads license plate numbers in diverse conditions for all types of vehicles.
Evidence Reporting
After a violation evidence image and vehicle license plate number have been captured, they are uploaded by the device to our servers along with corresponding information such as location, time of violation, class of vehicle, etc..
This is then forwarded to the traffic police authority's systems for automatic ticketing and subsequently collecting relevant fines & fees from the violator.
Violation reports generated automatically by our devices can also be accessed by the traffic police through a dashboard for manual verification & validation.
Current Progress
As of June 2023, we've completed the development of our initial prototype device. We are initiating an extensive real-world pilot programme to test the device in varying ground conditions and across multiple cycle lanes.
Depending on the outcome of this pilot programme, the Directorate of Urban Land Transport (DULT) will choose to deploy our devices in cities across the state. This would be a much-needed supplement to the state's initiatives for transitioning to sustainable mobility.
