Artificial Intelligence-enabled cameras installed near Mananchira, Kozhikode | Photo: Mathrubhumi
Kozhikode: The Safe Kerala project of Kerala Motor Vehicles Department has sparked a massive controversy in the state with the opposition Congress alleging corruption in the contract.
However, the debate on social media started with whether the cameras have artificial intelligence as claimed by MVD and the implementing agency, Keltron, a state-owned PSU. Let us try to understand what these cameras offer.
The 236-crore project included the installation of 726 cameras, setting up 13 control centres, including a state control centre, other related equipment, and maintenance of the centres and cameras for five years. Although it was reported widely as "726 AI cameras," not all of them have artificial intelligence.
Here is the breakdown:
- 675 AI traffic cameras
- 25 AI parking violation detection cameras
- 18 AI red-light violation detection cameras
- 4 speed violation detection cameras
- 4 vehicle-mount speed violation detection cameras
Also, it may be noted that it is often misreported that the cameras can calculate the average speed of a vehicle between two cameras and fine the owner by analysing whether the vehicle crossed the speed limit on the road. Keltron MD has stated that this facility is not available now. After all, there are only eight cameras for speed checks. A Keltron official told mathrubhumi.com that radar technology is used in speed check cameras, and AI cameras cannot provide accurate speed.
What is AI in this?
According to Keltron, the 675 cameras are equipped with specially designed artificial intelligence software (developed by Thiruvananthapuram-based Trois Infotech, according to Madhu Nambiar, MD of SRIT, a contract agency of the project engaged by the Keltron). Keltron also says that the cameras have a deep learning-based algorithm to detect traffic violations.
What is artificial intelligence?
AI refers to simulating human intelligence on machines, in other words, making machines think like humans. We all have been knowingly or unknowingly using AI these days. For instance, on Google Photos, searching for "red car" results in saved photos containing a red-coloured car. This represents a simple use of AI, according to Rahul (name changed on request), a Bengaluru-based engineer who works in the AI research and development wing of a prominent international company.
Deep learning is a subset of machine learning. Machine learning is “an application of AI that allows machines to extract knowledge from data and learn from it autonomously,” according to Google. One can feed a lot of data to the machine and ask it to learn from the examples. Deep learning is “algorithms based on highly complex neural networks that mimic the way a human brain works to detect patterns in large unstructured data sets.”
So, Keltron claims these cameras work on a highly complex network to process images of vehicles to check violations. According to a Keltron official who is working on this project, the camera is fed with "positive and negative images" for the learning purpose. Besides, the learning process will continue for at least five years during which the system will keep learning itself based on the input from the control centre.
"An automated feedback system has been established to inform the camera about the non-penalised notifications it sends," he added. AI has primarily two parts: training and inference. Following the training of the AI software, the system can infer based on the model. Rahul explained that these camera software systems are not as sophisticated as ChatGPT, which doesn't know what we're going to ask for, but the camera software knows what it is looking for.
The cameras have a hardware component developed by the American multinational IT company NVIDIA, based on their Edge Computing technology.
Rahul further explained that Edge AI is like a small computer, and it can process images within the cameras to provide real-time analysis. This eliminates the need to send images to a server for analysis, as the process is done within the machine itself. However, he added that the machine is costly.
“Through a combination of computing power, AI technology, data analytics, and advanced connectivity, the edge extends compute capabilities from data centres out to the edge of networks, allowing organisations to act quickly on data where it’s captured. Reducing the distance between where data is captured and where it’s processed not only alleviates data transit costs but also improves latency, bandwidth utilisation, and infrastructure costs,” notes the NVIDIA site.
As it contains a processor, it can even work without the internet. Since the use of cloud or server is not required, huge bandwidth or storage is not necessary to run it. Edge is being used in various systems worldwide for multiple purposes.
When asked if these violations can be checked without AI, Rahul said, "Yes, but the analysis will not be at this scale or real-time."
While NVIDIA provided the hardware, Trois Infotech coded the AI software to analyse traffic violations on Kerala roads. However, Trois Infotech phone numbers available on the internet were turned off when contacted, and they did not reply to the email query either.