Computer Vision in AI and its Types 🖥️👀
Learn the basics of Computer Vision and its types
In this blog we’ll cover what Computer Vision is and the different types of computer vision models. Computer vision is an evolving field of AI that has found it applications in a lot of things.
If you’ve ever wondered what Computer Vision is and wanted to know about it, just keep reading the blog to find out more. We will cover the three basic types of computer vision models i.e. Classification, Detection, and Segmentation. But first,
What is Computer Vision?
Computer vision is a field of Artificial Intelligence (AI) that allows computers to acquire, process and analyze digital images, videos and other visual inputs. The aim of Computer vision is to replicate and further improve upon the human visual system.
Computer vision is categorized into 3 broad categories:
Classification as the name suggests involves identifying the class of an object in an image. This classification is categorical in nature like “cat” or “dog”, binary output (0s or 1s) or other categories. In computer vision each classified output has a probability (accuracy of prediction) associated with it based on the performance of the computer vision model in classifying a given image.
The goal of classification is to group the output into a particular class. If the goal of our problem is to classify the data set into discrete or categorical classes then it is a classification problem.
In Detection the goal is identity where the object is in an image or input video. This is done by first classifying which object is present in the image and then drawing bounding boxes around where the object is located in the image.
Each bounding box drawn has a confidence score associated with it. A Computer Vision Engineer has the ability to tell the model to show only those bounding boxes that are above a certain confidence threshold while others are rejected.
The Detection process involves classification for identifying the object before drawing the bounding box around it.
Segmentation is about dividing the image into different subgroups based on similarities or differences in pixels characteristics to identify objects or set boundaries within an image. It helps reduce complexity and makes analysis of image easier.
Segmentation is further classified into:
1. Semantic Segmentation
It is a process in which a label is assigned to every pixel in the image. In this all instances of a class are considered as one.
2. Instance Segmentation
It is a process of detecting and localizing an object in an image. This process considers separate instances of a class as separated objects.
Applications of Computer vision:
- Computer vision is used extensively in humanoid robots to identify and interact with surroundings.
- Surveillance cameras & Traffic cameras to check people following traffic rules and identify bad actors.
- Drones use computer vision to identify enemy targets.
- Assembly lines to detect and remove defective products using captured image or video.
- Autonomous Vehicles or Self-driving cars to detect other cars or surrounding things.
- Supply chain, Amazon uses an army of robots using computer vision to pick and supply correct packages from warehouse to delivery.