How Edge Intelligence Is Driving the Next Wave of Computer Vision Applications in 2019

2018 marked a watershed moment for computer vision, with object detection and facial recognition accuracy reaching new heights. The proliferation of deep‑learning models—especially convolutional and recurrent neural networks—has delivered powerful, ready‑to‑deploy solutions. Yet these advances come with a price: the models grow in size and computational demand. For instance, YOLOv3, a widely used object detector, now contains 106 fully‑convolutional layers—more than twice the depth of its predecessor—while RetinaNet and SSD variants have similarly boosted performance at the cost of added complexity.

Keeping Up With New Demands

As camera deployments surge—particularly in high‑resolution, live‑stream scenarios—there is an urgent need to apply sophisticated vision algorithms in real time. Traditional passive surveillance no longer suffices; cameras must deliver proactive intelligence. Streaming full‑resolution video to the cloud is prohibitively expensive, bandwidth‑intensive, and latency‑prone. Conversely, deploying racks of high‑power servers on‑site consumes valuable space, electricity, and capital, especially when scaling across multiple locations. The challenge is clear: how can we process video from hundreds of cameras simultaneously with limited resources?

The Solution: Video at the Edge

The answer lies in edge computing. By embedding robust inference capabilities directly into cameras or by deploying compact edge appliances between cameras and the cloud, workloads can be distributed across many devices. This shift is supported by a growing ecosystem of low‑power, high‑performance AI processors. NVIDIA’s Jetson series delivers real‑time inference in embedded form, while Intel’s Myriad and Neural Compute Stick—products of its Movidius acquisition—offer similarly efficient solutions. New entrants such as Mythic and Graphcore have secured hundreds of millions in venture capital, and industry giants Google and Amazon have unveiled their own edge chips, underscoring the strategic importance of on‑device intelligence.

What’s to Come

Edge‑based vision will unlock a new generation of real‑time insights. Passive video recorders will evolve into active guardians—detecting children at risk of drowning, flagging weapons near schools, opening secure doors automatically, and identifying manufacturing defects. In retail, cameras will map shopper movements to optimize flow and reduce wait times, while safety compliance in factories will be monitored in real time. With over a billion cameras currently deployed and another billion on the horizon, edge processing is poised to make every camera truly intelligent.

Companies like Kogniz already provide video intelligence services that leverage edge appliances—standalone cameras and adapters for existing IP units—enabling on‑demand, infrastructure‑light deployment across unlimited cameras and sites.

Jed Putterman serves as the Co‑CEO of Kogniz. Mr. Putterman has founded multiple technology ventures, including Snapcentric (acquired by VeriSign) and Allerez (acquired by Mercury Interactive Corporation). He began his career at Oracle Corporation and consulted for major firms such as Sun Microsystems, SGI, and Aspect Communications. Mr. Putterman is a UC Berkeley alumnus.