Computer vision is a multidisciplinary field of study that teaches computers to interpret images and videos just like humans. The most
challenging area in computer vision is Object Detection which deals in
recognizing multiple objects in an image or video and classifying them
accordingly.
challenging area in computer vision is Object Detection which deals in
recognizing multiple objects in an image or video and classifying them
accordingly.
There are many applications of object detection, the most notable ones are
- Self Driven Cars, for detecting objects on the road.
- Security Surveillance, for keeping a watch in public
places for any untoward incidents. - Traffic Surveillance, to keep a watch on traffic
violations and accidents.
This article will deal with a specialized use case in traffic surveillance known as Automatic Number Plate Recognition. We will see what it is, learn about its technical details and how it can be used effectively to curb traffic violations.
What is Automatic Number Plate Recognition (ANPR) System
As the name suggests, ANPR is a system to capture the vehicle number by reading its number plate from traffic surveillance cameras. As you can see in the below illustration, the number plates of multiple cars have been detected and its number has been read accurately by the system.
Such software has existed since the 1970s, but over the years the underlying technology to recognize number plates has evolved and improved drastically.
A typical ANPR system has the following components:
- Computer vision and Optical Character Recognition to detect the number plates and read the number.
- Imaging Hardware, usually dedicated surveillance cameras installed on roads and highways.
- Database to store the vehicle information in case of any events like overspeeding, breaking red light, etc. that is detected by the ANPR system.
The most challenging part is, of course, the first part. Let us see in more detail how a computer can read a number plate from videos and images.
How ANPR works
Automatic number plate recognition is a 3 step process as explained below.
Number Plate Detection
In a traffic video, there would be multiple cars passing by every moment. So at first, it is very important to detect and localize the number plates of each car in the frame. This is essentially an object detection problem.
Number Segmentation
In this step, the number plate detected and localized in the above step is segmented into multiple regions to isolate each character in the number plate.. This is a very important step, because if the numbers are not segmented clearly then the system will fail to read the number.
Number Recognition
This is the final step, in which the characters segmented in the above step are classified as actual numbers.
Deep Learning Implementation of ANPR
Currently, all the computer vision problems show the best results when implemented with Deep Learning techniques, to be very precise by using convolutional neural networks.
At the moment the most powerful state of art deep learning technique for implementing ANPR is YOLO (You Look Only Once). YOLO algorithm is very fast compared to other algorithms and hence is suitable for the realtime use case of automatic number plate recognition with high accuracy.
How AeoLogic ANPR system is helping the Indian Government
Recently the Indian government came up with more strict policies to deal with traffic violations. AeoLogic ANPR system assisted the Indian government to leverage artificial intelligence to detect traffic violations and generate e-challans against the culprits by capturing their number plates.
AeoLogic ANPR system has been adopted by more than 18 Indian states and has helped traffic police to generate INR 50-70 Lakhs worth tickets per day. Gradually more states are expected to adopt this cutting edge artificial technology.
Conclusion
ANPR is used to read the number plates, but its underlying AI technology of object detection and optical character recognition can be used in other areas as well. It can be used for assisting blind people to make them aware of their environment or cashier-less stores like Amazon Go, where people can pick up any item, which is then billed by computer vision powered systems directly.