Book Image

Practical Computer Vision

By : Abhinav Dadhich
Book Image

Practical Computer Vision

By: Abhinav Dadhich

Overview of this book

In this book, you will find several recently proposed methods in various domains of computer vision. You will start by setting up the proper Python environment to work on practical applications. This includes setting up libraries such as OpenCV, TensorFlow, and Keras using Anaconda. Using these libraries, you'll start to understand the concepts of image transformation and filtering. You will find a detailed explanation of feature detectors such as FAST and ORB; you'll use them to find similar-looking objects. With an introduction to convolutional neural nets, you will learn how to build a deep neural net using Keras and how to use it to classify the Fashion-MNIST dataset. With regard to object detection, you will learn the implementation of a simple face detector as well as the workings of complex deep-learning-based object detectors such as Faster R-CNN and SSD using TensorFlow. You'll get started with semantic segmentation using FCN models and track objects with Deep SORT. Not only this, you will also use Visual SLAM techniques such as ORB-SLAM on a standard dataset. By the end of this book, you will have a firm understanding of the different computer vision techniques and how to apply them in your applications.
Table of Contents (12 chapters)

Image formation

The basic camera model is a pinhole camera, though the real-world cameras that we use are far more complex models. A pinhole camera is made up of a very small slit on a plane that allows the formation of an image as depicted in the following figure:

This camera converts a point in the physical world, often termed the real world, to a pixel on an image plane. The conversion follows the transformation of the three-dimensional coordinate to two-dimensional coordinates. Here in the image plane, the coordinates are denoted as where , Pi is any point on an image. In the physical world, the same point is denoted by , where Pw is any point in the physical world with a global reference frame.

Pi(x', y') and Pw(x, y, z) can be related as, for an ideal pin hole camera:

Here, f is focal length of the camera.

For further discussion on geometry of image formation...