By Tech Brew Staff
less than 3 min read
Definition:
Computer vision is a branch of artificial intelligence that allows computers to gain information from visual inputs, like video and images, and use that information to make decisions, much like how human vision works.
Computer vision involves training on reams of data and then using machine learning to analyze those inputs repeatedly until the computer recognizes the images.
Computer vision is the way autonomous vehicles “see” the road in a similar way to human drivers. AV systems use computer vision to understand visual inputs that come from the vehicle’s algorithms, cameras, radar, and other sensors, helping the vehicle identify things in the world around it, like other cars, pedestrians, traffic signs, cyclists, and more. The vehicle’s self-driving system then uses that information to make decisions in real time.
One of the challenges of scaling autonomous vehicles is that there are an endless number of driving scenarios that could happen in the real world. AV systems, using computer vision, AI, and other tools, must be able to safely and appropriately respond to any and all of them.
Autonomous vehicles use machine learning to analyze the data they gather from their cameras and sensors, and to interpret that data, find patterns in the visual inputs, and then make decisions about how the vehicle responds.
AVs use computer vision for functions like object detection and tracking, and analyzing data from lidar (light detection and ranging), according to computer vision library OpenCV.
“Self-driving cars use a mix of sensors, cameras, and smart algorithms to move around safely,” per OpenCV. “They need two main things to do this: computer vision and machine learning.”
AVs might use a “computer vision model to categorize every pixel in a high-resolution image” of a scene featuring numerous objects it might encounter in a real-world driving scenario, according to an MIT blog post about a 2023 development out of the MIT-IBM Watson AI Lab: “a more efficient computer vision model that vastly reduces the computational complexity of this task” and that can do this type of analysis in real time using the vehicle’s on-board computers.
