Computer Vision for Developing Social Distancing

With the advances in computer vision and machine learning, computers are now learning to track people’s movements. This technology has ramifications for social distancing, where we can tell when people are feeling uncomfortable. For example, one of the practical uses of this kind of technology is to detect stalking behavior.
Social distancing using computer vision

Social distancing is a vital strategy to avoid coronavirus. Several applications are available using computer vision to detect people’s count in the area. Unfortunately, we’re living in an age of coronavirus. This virus causes pneumonia and multi-organ failure and has a mortality rate of over 30%. This disease is devastating and can be very hard to track. It looks terrifying because coronavirus is transmitted through coughing or sneezing. 

Using a computer vision system to measure the distance between people could detect signs of illness. Monitoring people for sneezing and coughing will provide the first line of defense. Combining this with computer vision technology can help automate monitoring by simply looking for a visual cue, a person covering their face with a hand. This blog will explain the research behind computer vision applications for developing social distancing.

Grand View Research says the global computer vision market size was valued at USD 11.32 billion in 2020 and is expected to expand at a compound annual growth rate (CAGR) of 7.3% from 2021 to 2028.

Computer Vision Solutions for Developing Social Distancing Tools

People Detection Algorithm

People detection is a computer vision solution for getting body sensors to track people accurately, without explicit labeling. A related technique, social distancing, is used to avoid collisions using only sensors. This model combines these two techniques to give an algorithm for tracking people across spaces with varying floor plans and clutter levels. By solving the problem of tracking people through cluttered environments, we can also solve the problem of social distance indoors.

This algorithm is based on social network analysis/graph theory. It uses AI and data mining to infer and track targeted individuals and entities who pose a potential threat requiring social distance and applies the principle of social distancing via an alternate medium or location to mitigate any risk.

It is a system for detecting and tracking people moving through physical spaces. We use it as a back end to our Real-time tracking software. It takes the image from a camera, extracts a track from that image, and then matches it against data from previous images. The result is a set of measures that show confidence that the current subject is still in view or no longer in view.

For instance, in a manufacturing company, the people detection algorithm can detect different objects and people.

People Detection Algorithm

Source: Always AI

Centroid Tracking Algorithm

A centroid tracking algorithm is one of the most commonly used computer vision approaches to social distance. The basic idea behind social distance is to keep people apart during an emergency without worrying about bumping into each other or impeding traffic flow. So, for example, you might want to keep people away from an area affected by Covid 19.

It uses video input and targets to find the targets’ centroids very quickly. Every step gives the coordinates of the current centroid and estimates where the centroids will be at the next step. Finally, it provides a prediction of where to look for them. If any of these predictions lie outside an exclusion area around the user, then that area is visible, otherwise not.

It’s a classic socialist strategy for social distance, implemented to keep track of the center of impact of the user base with the geographic center. As more people added, the centroid tracking algorithm, the variation in radius will decrease, leading to a higher population density. 

It’s a simple algorithm using the Least-Squares criterion, which defines the size of the region of space that a moving object will occupy. So that objects inside this region can stay in touch through motion. 

The algorithm is very efficient. Mathematically it closes the ‘communication circle’ for all of its members, keeping them connected despite any possible moves of the target object. 

Moreover, this algorithm performs when continuously moving target objects (in two or three dimensions) need to stay in touch with any stationary object within a bounded area (a certain number of objects may be allowed in this area), considering each other as potential targets.

For Instance, the centroid tracking approach is perhaps one of the modern methods for identifying human gatherings for social distancing purposes. As a result, several agencies have implemented the model, including the US Federal Aviation Administration (FAA) and NASA, among others.

Centroid Tracking Algorithm

Source: Science Direct

AI Vision-Based Social Distancing Detection Projects

Single Shot object Detection

Single Shot object detection can detect the presence of an object. Unlike Other Approaches, an SSD network programmed relatively few examples and no background knowledge needed, which requires large numbers of labeled samples for training. Instead, it is about running an image through a deep learning neural network and quickly detecting the distance between people.

In another way, Single Shot object Detection (SSD) can achieve social distancing by measuring the distance between persons, bounding box coordinates, and person’s depth using MobileNet and OpenCV.

Single Shot object Detection

Source: Towards Data Science

Yolo Algorithm

The Yolo algorithm detects objects in images, like faces. It does this by optimizing the difference between the input and reference images to make it extremely easy to find this difference.

It’s a machine-learning algorithm designed for automatically calculating social distances between people. The Yolo algorithm is the world’s first fully automatic social distance calculator. It can be integrated seamlessly into your business, making it easy to determine the distance between two people.

For instance, the Yolo object detection plays the leading role in detecting people in video streams.

Yolo Algorithm

Source: PY Image Search

Conclusion 

Social distancing captures the social impact of an environment. It calculates using empirical data collected through sensors—usually, cameras or audio recorders installed in a given place. 

Can measure the impact of terrorist attacks inside crowded places, now used for a broader range of use cases, including shops and malls security, retail surveillance, public safety reporting, airports safety reporting, and many others. The idea behind social distance is simple: measure the likelihood that two people interact or influence each other inside a particular environment or context. 

Visionify’s People detection and tracking is a computer vision solution for automatically detecting, tracking, and classifying people from video. The software consists of many frames captured from cameras or stored media from the live stream. It’s fast, accurate, and can work on any image. Call us to get a live demo.