The rise of artificial intelligence to fight COVID-19

The COVID-19 pandemic has significantly impacted the way we live, work and move around. The world came to a screeching halt with governments across the globe imposing lockdowns to prevent the spread of the virus. After months of standstill, factories and offices are slowly getting back to business armed with three weapons—social distancing, face masks and temperature checks.

For the Aditya Birla Group, safety of its employees has been of utmost importance in this severe health crisis. The Group turned to artificial intelligence (AI) and deep learning to keep its workforce safe at their manufacturing units and offices. The Group Data and Analytics cell, the Group’s analytics and artificial intelligence arm, built an efficient and economic approach by using AI to create a safe environment in both manufacturing setups and offices.

Relying on the already proven technology, VEDA (Video Enabled Decision and Alerts), the team built a refreshed version with advanced capabilities to cater to the needs of the manufacturing units and offices post COVID. “VEDA, an AI platform and a first-of-its-kind proprietary product, uses computer vision technique to provide alerts and insights to take real-time decisions,” explains Deep Thomas, Group Chief Data and Analytics Officer, Aditya Birla Group.

The VEDA platform deploys custom-built artificial intelligence/machine learning (AI/ML) models for analysing videos and providing customised solutions for solving business problems. “VEDA was an idea that emerged from a hackathon in 2019. Six months later it was successfully deployed at a plant in Taloja. During the lockdown, we added new features to the platform that enhanced worker safety in the post-COVID world,” adds Naveen Xavier, Vice President, Group Data & Analytics, Aditya Birla Group. The platform is scalable and built for quick deployment to process multiple video feeds in parallel, apply advanced video analytics in real time, build on the model output through configurable business logic and raise real time alerts, notifications and reports.

Refreshed VEDA: Revolutionising manufacturing and workspace

During the current pandemic, the team identified artificial intelligence as a core tool for real-time safety analysis at factories and offices and developed and deployed a refreshed version of VEDA with three new and relevant features for the post-COVID world.

Temperature monitoring

Fever being a symptom of a corona virus infection, temperature checks at office entrances have become essential to prevent the spread of the virus. However, using handheld IR sensors for temperature checks of a large workforce can be a time-consuming process. To make this process more efficient, the Group is using VEDA by integrating it with cameras that provide dual feeds—the RGB feed from the CMOS sensor and the thermal feed from the VOX (Vanadium Oxide) sensor.

Face detection happens using the RGB feed and temperature check is done from the ROI of the forehead that is visible above the mask. Thermal cameras read the temperature with an accuracy of +/- 0.5oC at a distance of two metres from the camera. If a temperature reading beyond the defined threshold is detected, an alert in the form of an audio signal is triggered which enables security personnel at the entrance to take suitable measures.

Facial mask detection

The World Health Organisation (WHO) recommends wearing face masks to curb the spread of the virus. To avoid manual checks and safeguard our security personnel, our team equipped VEDA with contactless detection of face masks to ensure that face masks are worn at all times as per guidelines. However, face masks are not standardised. For detecting face mask, a two-step process has been undertaken—face detection and mask classification. For face detection, VEDA already uses deep learning-based CNN models. In order to train the mask detection model, the face detection model was used to extract face-crops from video feeds. These face crops were manually annotated into mask and no mask categories. A rudimentary deep learning off-the-shelf CNN classifier was built on the dataset. Next, the face detection model on the video feeds from multiple plants and from cameras ranging at different heights were analysed. The captured images were then trained through a deep learning binary classification model to detect facial masks.

Adherence to social distancing

Besides temperature monitoring and face mask detection, a robust social distancing measurement algorithm has also been built into VEDA. The algorithm is a mix of modern-day deep learning and classic projective geometry techniques. The VEDA platform has the social distancing monitoring module residing as part of the model layer in the solution. When any video feed records recurring violations lower than the threshold distance value for minimum specified duration, an alert is triggered. The module detects workers in a frame and then calculates the distance between each worker. The team used the transfer learning approach and fine-tuned MobileNetV2 model for person detection using TensorFlow framework. Using the person detection model on each frame, the platform provides the output from the model in the form of bounding rectangles on detected persons to detect the distance. To find out if the detected people are following social distancing, backend calculation takes place. Accuracy of the distance is based on transformation matrix, which in turn depends on accurate calibration of ground points.

The refreshed version of VEDA is currently integrated with features to monitor worker safety such as usage of face masks and maintaining social distance to manage safety requirements during COVID restrictions. Our manufacturing units and offices have already deployed VEDA for COVID-related safety measures. “VEDA is a powerful platform. We used it successfully for changing people’s behaviour, attitude and habits in the manufacturing plant,” says Vaishali Surawar, Unit Head, Hindalco Industries Ltd at Taloja. In addition, VEDA can also monitor usage of PPE gear such as helmets and vests, enhancing fire detection with visual surveillance, alerting on restricted area intrusion, vehicle detection, visual quality inspection and raise real-time audio alerts with incident trends being captured on a dashboard.

The VEDA platform is highly configurable to address various use cases through video analytics and can be easily scaled across plants and workplaces through on-premise deployment to drive benefits in the areas of safety, automation and reliable operations where automated visual surveillance is required. The implementation of video analytics will aid transformation in a business environment by increasing process efficiencies, understanding customer behaviour and enhancing safety in diverse environments while considerably reducing manual intervention. VEDA delivers intangible amount of cost benefits by monitoring fire, oil spillage, crowd formation, people movement in unsafe zones and use of safety gear by workers on shop floors.

Based on cutting edge technologies, the VEDA platform offers an efficient and economic solution for enhancing people safety and workplace security.