Book Image

OpenCV 4 with Python Blueprints - Second Edition

By : Dr. Menua Gevorgyan, Arsen Mamikonyan, Michael Beyeler
Book Image

OpenCV 4 with Python Blueprints - Second Edition

By: Dr. Menua Gevorgyan, Arsen Mamikonyan, Michael Beyeler

Overview of this book

OpenCV is a native cross-platform C++ library for computer vision, machine learning, and image processing. It is increasingly being adopted in Python for development. This book will get you hands-on with a wide range of intermediate to advanced projects using the latest version of the framework and language, OpenCV 4 and Python 3.8, instead of only covering the core concepts of OpenCV in theoretical lessons. This updated second edition will guide you through working on independent hands-on projects that focus on essential OpenCV concepts such as image processing, object detection, image manipulation, object tracking, and 3D scene reconstruction, in addition to statistical learning and neural networks. You’ll begin with concepts such as image filters, Kinect depth sensor, and feature matching. As you advance, you’ll not only get hands-on with reconstructing and visualizing a scene in 3D but also learn to track visually salient objects. The book will help you further build on your skills by demonstrating how to recognize traffic signs and emotions on faces. Later, you’ll understand how to align images, and detect and track objects using neural networks. By the end of this OpenCV Python book, you’ll have gained hands-on experience and become proficient at developing advanced computer vision apps according to specific business needs.
Table of Contents (14 chapters)
11
Profiling and Accelerating Your Apps
12
Setting Up a Docker Container

Implementing a Sort tracker

The Sort algorithm is a simple yet robust real-time tracking algorithm for the multiple-object tracking of detected objects in video sequences. The algorithm has a mechanism to associate detections and trackers that results in a maximum of one detection box for each tracked object.

For each tracked object, the algorithm creates an instance of a single object-tracking class. Based on physical principles such as an object cannot rapidly change size or speed, the class instance can predict the feature location of the object and maintain tracking from frame to frame. The latter is achieved with the help of the Kalman filter.

We import the modules that we will use in the implementation of the algorithm as follows:

import numpy as np
from scipy.optimize import linear_sum_assignment
from typing import Tuple
import cv2

As usual, the main dependencies are numpy...