by Naveen Gattu, COO and co-founder of Gramener
Retail stores, libraries, museums, banks, and stadiums consume significant customer traffic every day. The business users are not concerned only about tracking their customers. Their safety is also an utmost priority. However, it is being proactively dealt with by placing cameras.
Yes, a camera saves the day. Cameras capture the movement of people in a structure and estimates almost accurate footfalls. The amalgamation of camera trap data with Artificial Intelligence technologies such as deep learning and Convolutional Neural Network gives rise to business-friendly use-cases of crowd counting.
Crowd counting is a niche technology that has skyrocketed in recent times. Research predicts this market to reach a whopping $1100 Million by 2022. Giants and SMEs from Retail and Event Management segments can use this technology to analyse their ROIs like never before.
What is Crowd counting AI model?
A crowd counting system is an AI-enabled smart system. It gathers data, which has the ability to offer actionable insights to users.
For example, the data helps retail managers to figure out their conversion rates. A retail store manager can count the customer footfalls at different sections of the store. With this, he can know the peak hours of customer visits. It helps them to allocate staff to attend customers, which further creates value in terms of revenue, profit, and customer experience.
Counting anything in the world with A.I. models.
The benefits of crowd counting applications are not limited to commercial businesses. Many not-for-profit organizations also utilize technology to help save endangered species. For example, Microsoft has collaborated with Gramener to count the penguin population in Antarctica. The objective is to count the sentinels of Antarctica from images taken by the camera traps installed at various locations, aided by deep learning-powered crowd counting models. Variation in penguin population can result in larger changes to the dynamic Antarctic ecosystem, so scientists are working on measuring these changes yearly to understand how changes in an ecosystem can impact the dynamics of penguin habitats.
Another example is the Washington-based Nisqually River Foundation, which counts salmon fish automatically using AI-enabled deep learning models. It is saving expert biologists hundreds of hours that they had previously been spending doing the work manually.
Actual count vs. Machine predicted count using heatmap technique
Benefits of Deep Learning over traditional models.
Deep Learning has evolved significantly in the past decade and is widely being applied in use-cases that include pictures, videos, and audios. Deep Learning techniques like Convolutional Neural Networks (CNN) are overtaking the traditional detection and regression-based models to offer better crowd counting use cases.
CNN is a subdomain of deep learning and a class of deep neural networks used to analyze image data. For example, in the case of automatic face recognition, deep learning models digest thousands of facial images and identify the important features by itself. These include features which humans might not notice, like wrinkles, the curvature of the cheek or the shape of the nose. It helps the machine to predict gender and age from the captured image.
Deep learning enabled crowd counting AI models can overcome several disadvantages that traditional regression models face. A few are:
- Tackling occlusion or locating people who are hidden from the view.
- Density differences where, in one frame there are several clusters of people, large and small.
- Perspective distortion where faces at the front of a crowd seem bigger than those at the farther end of the crowd.
- Camera angle where different images from different angles of the camera need different training.
Businesses benefit from crowd counting.
Deep learning is rapidly pushing computer vision to a stage where businesses can exploit its benefits integrating models in everyday use.
Retail Industry: Understanding customer traffic is important in the retail industry. It helps them organize their merchandise and optimize store layouts. Crowd counting AI model has great potential in the retail industry. The retail managers can count varying store footfalls every day. Using the crowd data, they can estimate conversions and measure the success of their campaigns.
For example, a retail store starts a sale in its cosmetic section. They launch two campaigns of different discounts for a week each. After the sale is over, the retail store manager can count the customer visits for both weeks and understand the success of both campaigns. That’s how crowd counting AI models can induce data-driven decision making in a retail business.
Event Management Industry: Similarly, the Event Management industry hinges largely on the data of the event attendees. A reliable estimate of the crowd can be a true marker for positive returns on their investments. Public events such as communal festivals, fundraisers, and political rallies are usually not ticketed. Crowd counting AI models can help understand the success of such events by noting the footfalls.
Also, it can help us identify the time frames where the crowd is more or less. This defines the part of events which are interesting or boring. Public events result in human intervention on a larger scale, which raises security concerns. Keeping a track of all the attendees mitigates this risk to a greater extent.
No matter what the venue is, a count of people you bring into your space is a valuable insight for planning future events.
Pharmaceutical Industry: Pharmaceutical industry is one of the niche industries which hardly takes any help from A.I. But, crowd counting algorithms has gained significant importance for drug characterization during the drug discovery process.
For example, we tackled a biological cell counting problem using crowd counting model for a pharma client. The scientists wanted to count the cells in various microscopic images for new drug discovery. For starters, the process is manual and painstaking. Furthermore, the biological cells in a microscopic image don’t have perfect shapes. While counted manually, any shape close to a circle (e.g. sphere) is counted to be the same cell type. The automated cell counting in the drug discovery process can reduce the manual efforts of the scientists. They can dedicate their time to other important works which suit their paygrade.
Journalism: 2013 saw one of the biggest uprisings in Egypt’s history when the citizens protested on the streets against their elected President, Mohammed Morsi. The estimation of the crowd by journalists and News channels was approximate. For some, the crowd was of 15 million and for others, it was 30 million. No one knew the exact number of protestors. Crowd counting algorithms can answer such questions and news channels can create huge viewership with true facts. A crowd counting application can estimate or count the number of people in an image or video.
Conclusion.
Crowd Counting AI model is a budding technology with a plethora of use cases in core domains. Retail, event, pharma, media, hospitality and many other industries can fruitfully benefit from its business applications. Unidirectional people counting system is expected to grow by $600 million by 2022. Amidst of all the research and forecast, the young entrepreneurs need to make data-driven decisions to grow their business.
Naveen Gattu is the COO and Co-founder of Gramener, a data science company. He is a proud recipient of the Lufthansa Pioneering Spirit Award and a good relationship and team builder. Naveen advises large enterprises in Data Driven Leadership strategies/roadmap and is a regular speaker at Industry/academic forums like Open Data Conference, Nasscom, TiE, Big Data Conferences. He also advises entrepreneurs/tech founders at Founders Institute, Startup Leadership Programs (SLP), Startup Sundays etc.
