Indian-American tech company Calsoft, in partnership with Nvidia , has developed an automatic number plate recognition (ANPR) system to help automate tollbooths across India. Vipin Shankar - Senior VP Technology - Engineering at Calsoft, said ANPR can also enhance traffic flow, enable pro-rated charging, improve parking management, detect fake number plates, monitor average speeds and aid in theft detection and rapid recovery.
The TomTom Traffic Index 2023 highlighted Bengaluru and Pune as the sixth and seventh most congested cities globally, with drivers spending an average of about 28 and 27 minutes, respectively, to travel 10 kilometres.
“Using the ANPR system in India can ensure smooth-flowing traffic on highways, leading to reduced travel time and lower consumption of fuels resulting in saving precious forex,” said Shankar, adding, “We intended to build a system to solve these problems on Indian roads, especially highways to help smooth traffic. For this, we used Nvidia’s technologies.”
Suite of tech
To automate tollbooths across India, Calsoft deployed a suite of Nvidia technologies integrated with the country’s leading payment system, the Unified Payments Interface (UPI), for a client. The solution could automatically read passing vehicle plates and charge the associated driver’s UPI account.
“While traveling on a highway, you pay one flat fee. With this technology, cameras can detect entry and exit points. Based on this, it can charge you on a prorated basis. This will save money. That’s one advantage,” he added.
This technology uses Nvidia Metropolis, an application framework, a set of developer tools and a partner ecosystem that brings visual data and AI together to improve operational efficiency and safety across industries. Alongside, the company’s engineers used Nvidia Triton to deploy and manage their AI models and the Nvidia DeepStream software development kit to build a real-time streaming platform. With this, data streams can be processed and analysed; capabilities like real-time object detection and classification are incorporated.
Calsoft also used NVIDIA hardware, including Nvidia Jetson edge AI modules and Nvidia A100 Tensor Core GPUs in its AI solutions.
Challenges in system
In India, challenges surrounding the license plate system include non-standard formats, non-uniformity of the background in terms of colours and size of text, multi-lingual or vernacular number plates and environmental factors like fog, snow and rain. Additionally, the varying locations of plates on vehicles and non-reflective materials at night complicate readability.
Shankar explained, “One challenge in the implementation stage was capturing the right image. Low-resolution cameras would not cut it. Initially, only around 40 per cent of the number plates were recognised correctly. With good cameras, we also got good quality images. Then, we improved the algorithms by training them on thousands of different language images and format images. Today we have hit around 95 per cent accuracy.”
He said this specific application of ANPR, which the company designed for the Indian market, addresses said challenges using AI/ML, specifically, deep learning (DL). “Within that, we use deep neural network (DNN) techniques and you only look once (YOLO), an object detection algorithm coupled with some Nvidia Metropolis microservices. Some of these ML algorithms and techniques can amplify low-light images, remove the noise in the images,and dehaze them,” Shankar said.
Around 1.5 lakh road traffic deaths were recorded in 2022 in India, with almost 60 per cent of them due to over-speeding, he added. ANPR can monitor vehicles’ average speed and identify those exceeding the limit.
“A fine ticket can be generated between two camera points immediately and sent to the car driver who can be warned to minimise the chances of accidents,” he said. If a vehicle is going beyond the allowed speed limit, a ticket can be issued to the driver online along with an alert to slow down. Alongside, the police can be informed of speeding.
Parking management in convention centres and other areas that see an inflow of a lot of vehicles can also benefit from this technology. Registration of vehicles can allow pre-approved or known vehicles to park free or at discounted rates, with non-registered vehicles charged per hour basis. Integration with mobile phones can facilitate direct payment and pre-booking can be done with cloud-based systems at venues. The technology can also detect fake number plates and help track stolen cars.