India is a land of diverse cultures and festivals are an inherent part of the cultural fabric of this nation. While they bring with them much fanfare and joy, festive gatherings usually teem with crowds that are often difficult to control. The inability to manage tens of thousands of people lead to accidents, and in extreme cases, stampedes and mass casualties. 

Maintaining law and order is a critical aspect of governance, and technology is necessary to to manage huge crowds with ease. It is this very challenge that the Emerging Technologies Wing’ under the Information Technology, Electronics and Communications Department (ITE & C) aims to resolve under its first-of-its-kind dedicated initiative. Under its umbrella falls the vision to develop a conducive ecosystem for emerging technologies in Telangana. The idea is to leverage emerging technologies to enable better governance and improve citizen service delivery. 

Over the last two years, the ITE & CE department has worked extensively across technologies, including AI, Blockchain, Drones, Cloud, Cybersecurity and E-waste. In fact, 2020 was declared as the ‘Year of AI’ in Telangana, to accelerate the state’s AI innovation ecosystem. The state-specific AI framework was released in June 2020, to identify the required initiatives for ecosystem development. Consequently, the Telangana AI Mission (T-AIM) was established to undertake the implementation. 

Delving deeper into the problem

Picture this - close to 11 crore people visit the city of Haridwar during the Kumbh Mela. This statistic is almost twice the population of the United Kingdom, and more than four times the population of Australia. 

Similarly, in Mumbai, more than 20 lakh people participate in the Ganpati Visarjan, every year. That’s not all - more than a crore people from six states attend the Medaram Jathara in Telangana. 

The situation is unmanageable, to say the least - it isn’t possible to predict the number of people based on the images, leave alone maintaining constant vigil. Moreover, it is hard to predict erratic crowd behaviour, and at the same time, difficult to uniformly deploy manpower, as crowd densities fluctuate over time. Plus, there is evidence to prove that unmanaged crowds often get agitated and pose safety threats, especially in the form of stampedes. 

Out of 400+ startups, Awiros was selected by the Telangana government for implementing AI applications in the state. The startup has developed an array of AI-based applications for better governance and safety of citizens, the first of which was an AI-based Crowd Estimation and Management application. What’s more, they have developed the world’s first operating system for the deployment and management of video AI apps. In due course of time, they have worked with the Telangana government on several large-scale projects. 

Today, their solution caters largely to the police departments. The Telangana police have used it effectively to manage crowds at the India vs West Indies T20 match in Hyderabad in 2019 as well as the Medaram Jatara Festival in 2020. 

Cracking the Crowd Conundrum

Here’s how the problem is solved, using a step-by-step procedure. To begin with, there is estimation of large crowds, moving on to the prediction of crowd density, and identification of bottlenecks, all using deep learning. The videos are first analysed using surveillance cameras, after which the analysis of crowd variation is assessed. Last but not the least, the areas of high crowding probability are identified. Thereafter, action is taken through strategies like manpower deployment and public announcement. There is also rerouting crowds to less crowded areas. 

The solution is the first application of its kind that can estimate crowd densities in real-time using surveillance cameras. Not only does this help the authorities detect the problem, but also tracks the variation in crowd density in an area to facilitate corrective measures so that untoward incidents are averted. 

Let’s get a better understanding of the solution. It utilizes camera feed from a network of surveillance cameras. The video stream is fetched from the Video Management System (VMS) server and is then analysed using Awiros Video Intelligence Engyn that runs on the Awiros operating system. 

Using a state-of-the-art deep learning-based object classifier, human figures are identified in the camera frame to count the number. This helps to determine crowd density, and in case this parameter exceeds the threshold defined by the user, then alerts are sent out. This solution is versatile and can be deployed On-Premises, Cloud and Edge+ Central configurations, and can be integrated with every component of a surveillance infrastructure, including Video Management Systems (VMSs) and Integrated Command and Control Centre (ICCC). 

A few minor challenges

As with everything, there are slight technical challenges in its functioning. A case in point is the India vs West Indies T20 match, during which the app was deployed on a PTZ camera positioned on the east-side stand in the stadium. The camera was then rotated to focus on each of the spectators’ stands in the stadium, one after the other. Every single time, the app was used to estimate the number of people present in the stands. These estimations were then added to figure out a final count of the total number of people in the stadium during the duration of the match. Despite this, a crowd of 25,000+ was estimated in the stadium, with an accuracy of 93%. 

Similarly, at Medaram Jatara, Awiros AI-based Crowd Estimation and Management System was installed on three different cameras on the premises, to estimate crowd densities as well as track the variation in crowd densities in real-time, throughout the duration of the festival. 

The impact and scale of this solution is huge, and can be compared through these numbers. Although only three cameras were used through AI, the standard monitoring utilised 12,000 police personnel, 350 CCTV cameras, 20 special cameras, and more.

All in all, this one-of-a-kind solution holds a competitive advantage like no other. It can estimate crowd densities in real-time, generate information and alerts both on desktop and mobile, and boasts high compatibility. Last but not the least, given the adequate infrastructure, the same solution can be scaled for any number of video feeds. 

Managing crowds effectively has been a pain point for law enforcement officials for long. Fortunately, this solution makes it a breeze!

Image by Tourism Victoria via Flickr