AI • IOT • NETWORK EDGE

Every driver knows how hard it is to keep their eyes on the road when tired—and how easy it is to become distracted by a text message, radio dial, or cup of steaming hot coffee. For professionals, who spend far more hours behind the wheel than the rest of us, staying focused while driving is even more of a challenge.

But now, emerging Advanced Driver Assistance Systems (ADAS) based on edge AI and computer vision help solve the problem of fatigued and distracted driving in ways that traditional solutions cannot. That’s good news for everyone—and a relief for fleet management, logistics, and ride-hailing businesses.

“Distracted and fatigued driving are major concerns for enterprise safety officers,” says Srini Chilukuri, Founder and CEO of TensorGo Software Pvt Ltd., a platform-as-a-service provider focused on computer vision and deep learning solutions. “ADAS solutions use edge AI to improve on older safety systems, offering real-time monitoring, analysis, and alerts to help drivers to focus.”

And while deploying AI solutions at the edge is challenging, partnerships between computer vision specialists and hardware manufacturers help get these innovative systems into commercial vehicles and on the road.

Edge AI on a Raspberry Pi

Case in point is TensorGo’s work with Intel on its Advanced Driver Attention Metrics (ADAMS) solution. The ADAS system design is elegantly straightforward: It comprises a compact camera, an edge computing device, and computer vision algorithms that monitor for risky driving.

ADAMS runs three separate AI behavioral detection algorithms concurrently:

• Drowsiness detection analyzes the driver’s face for signs of sleepiness, such as frequent yawning or closing eyes.
• Head pose picks up on distracted driving by identifying instances of drivers looking away from the road, such as adjusting the navigation system or reaching for a dropped item.
• Object detection spots when a person is glancing at a distraction such as a cell phone.

If any of the algorithms detect a problem, the system immediately alerts the driver via their mobile device and then sends a second alert to a company safety official as well.

Although the basic system architecture was established in the product development phase, bringing a working version of ADAMS to market presented challenges. The proof-of-concept ran on a bulky edge device that ultimately proved too inefficient and inflexible to turn into a viable product. TensorGo’s engineers wanted to migrate their system to a compact and energy-efficient 32-bit Raspberry Pi edge device and a Raspberry Pi camera. But it wasn’t clear how it would be possible to run multiple AI algorithms on a smaller edge device without overtaxing the processor.

Working with Intel, the TensorGo team overcame their engineering challenges. They used the Intel^® OpenVINO™ toolkit to optimize and accelerate the AI algorithms to run efficiently on the compact Raspberry Pi device. Intel architects also suggested a strategy of processing fewer frames of camera video data than in the original prototype. This approach provided more than enough data for high-precision computer vision analysis—while also reducing the burden on the processor, thus improving ADAMS’ overall performance and stability.

Case Study Shows Improved Safety—and Cost Savings

TensorGo’s deployment with a large trucking and delivery company with operations in the Middle East demonstrates the capabilities of ADAS systems in real-world scenarios.

The company was facing an increasing number of accidents across their fleet of more than 500 trucks—with driver distraction and fatigue being identified as the main cause. Management could not accept the safety risk to drivers and the general public. They were also concerned about operational efficiency issues due to vehicle downtime and liability costs. Despite implementing driver training programs, the problem persisted.

Working with TensorGo, the company deployed an ADAMS system in every vehicle in their fleet. Within six months, the results were conclusive—the edge AI approach was a resounding success. The company saw a 32% reduction in distraction-related incidents and a 27% decrease in fatigue-related accidents. The driver attention system had also helped improve on-time delivery rates by 18%, leading to an estimated cost savings of more than $1.5 million.

“ADAS systems like ADAMS are a game changer for enterprise safety officials,” says Chilukuri. “They improve safety outcomes and positively impact the bottom line, solving key safety challenges and helping to overcome adoption barriers.”

By combining powerful #safety and cost savings benefits, #ADAS solutions are an attractive option for #FleetManagement companies. TensorGo via @insightdottech

The Future of Transportation Safety and Beyond

By combining powerful safety and cost savings benefits, ADAS solutions are an attractive option for fleet management companies, leading to an increased uptake of these systems over the coming years.

TensorGo is preparing for this future with plans to introduce more features to its existing solution. The company is looking at ways to add a GSM module to ADAMS so that alerts can be emitted directly from the edge device rather than the driver’s phone. The engineering team is also exploring how to incorporate AI collision detection models into their solution to alert drivers to potential road hazards.

Beyond ADAS systems, the solution’s underlying technology can support other use cases. The core software and computer vision technology used in ADAMS can be adapted to applications including workplace safety, assisted living monitoring, and industrial operations.

“AI and computer vision at the edge will play a transformative role in, logistics, and other sectors over the coming years,” says Chilukuri. “Real-time monitoring and analysis will improve safety and efficiency across the board, and we aim to be a key player in that transformation.”
 

This article was edited by Georganne Benesch, Editorial Director for insight.tech