This article is part of TechXchange: AI on the Edge
What you’ll learn:
- How edge AI is fostering a new era of intelligent video analytics (IVA) with myriad new capabilities and possible uses.
- The impact of enhanced processing power of new edge AI technology on AI-assisted video devices across different industries, affecting both the quality and quantity of the tasks achievable with IVA devices.
Intelligent video analytics (IVA) are on the rise across numerous industries thanks to recent developments in edge AI technologies. AI-integrated analytics solutions abound—devices such as intelligent cameras, intelligent network video recorders (NVRs), and edge AI boxes are now capable of powerful on-device AI processing.
The ability for cameras and other small devices to identify objects, track movement, and even recognize behavior first arrived at the edge in a limited capacity, providing unreliable alerts and tracking as well as low-quality detections. But today’s AI processors can bring exponentially more teraoperations per second (TOPS) to existing cameras and devices and run state-of-the-art neural-network (NN) models at thousands of frames per second (fps). That kind of computing power now available to these edge devices was once exclusively reserved for the cloud.
For proof that mid-range edge AI solutions are entering the mainstream, look at a recent study conducted by Omdia research commissioned by Hailo. The report estimates that in 2021, 26% of cameras and NVRs sold were AI-capable, and predicts that by 2025, AI-capable cameras will comprise 64% of worldwide IP camera sales.
The Impact of Enhanced AI
The enhanced processing power and efficiency of new edge AI technology has brought tangible improvements to AI-assisted video devices. It impacts both the quality and quantity of the tasks achievable with these devices.
With lower latency and higher frame rates, deep-learning-based tasks can run on edge devices in real-time. Real-time frame-rate processing allows for more responsive interfaces and is particularly crucial for tasks that must detect and track fast moving objects.
Edge AI also affords high-resolution cameras (UHD, FHD) the large number of TOPS needed to accurately detect small details and provide actionable image analytics. More efficient AI processing power lets AI applications use more complex and therefore more accurate NN models, resulting in better detections and classifications and more reliable alerts.
System-level cost savings represent another enticing by-product of edge AI improvements, as more powerful on-camera IVAs can cover more regions of interest (RoI) with less impact on a company’s bottom line. Rich metadata and analytic insights can be easily tracked, recorded, and transmitted without the need for pricey on-device storage or cloud-based processing.
Quality Assurance Automation
Quality assurance is as important as it is challenging, particularly for manufacturing and industrial processes. Some research estimates that defects can cause 50% of production to go to waste, a figure that rises up to 90% with more complex manufacturing lines.
Recent innovations have produced highly accurate machine vision that scans batches of product for defects at a volume and level of detail far beyond what the human eye can handle. However, the data-processing capabilities needed for completely automated inspection need to be local—on the production line. This is due to strict latency limits, which are tied to the conveyor belt or other production equipment that operate continuously.
Powerful edge AI can support simultaneous analysis of very-high-frame-rate cameras and multiple streaming camera inputs in real-time, allowing for more swift, detailed factory monitoring and product inspection. Rates of production error will drop, saving factories time, money, and material.
As AI-powered computer processing improves, visual intelligence and sensory perception will quickly become the new standard for Industry 4.0.
The Analytics of Retail Security
Line-crossing detection has long been the primary method for monitoring footfall of incoming and outgoing shoppers, but it’s not always reliable. AI-integrated cameras allow for more accurate tracking with little to no missed counts by using blind-spot recovery and depth estimation. AI devices also can support automated entry-point detection, which is particularly useful for stores or retail spaces where the entry point is dynamic or not a clearly defined doorway.
Furthermore, AI cameras can provide business intelligence (BI) insights for retailers. What parts of the store attract customers the most? What parts of a store made the most people stop? Intelligent cameras can shed light on these questions and help business strategists derive customer-intent projections or traffic heatmaps, enabling them to reconfigure store layout or displays to maximize profits and enhance the shopping experience. In addition, loss prevention can be bolstered through AI identification of vulnerable or frequently stolen products.
The Future of IVA
The robust processing abilities of edge AI are shifting IVA from simple object and motion detection to full-fledged scene understanding and context awareness. The former offers a metadata-rich view of a given ROI, while the latter provides a perception of a single detected object within the spatial and temporal context of a video recording.
As scene-understanding and context-awareness processes continually improve, these analytics devices will increasingly be able to integrate intent analysis and even behavioral analysis into their host of applications.
Read more articles in TechXchange: AI on the Edge