Improved AdaBoost Algorithm and Object Recognition Based on Haar-like Training
Li, Xiuyang (2016)
Vaasan ammattikorkeakoulu
2016
All rights reserved
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2016060411894
https://urn.fi/URN:NBN:fi:amk-2016060411894
Tiivistelmä
The main aim of this thesis is to introduce a new improved AdaBoost algorithm based on the traditional AdaBoost algorithm of improving the accuracy and speed of traditional AdaBoost algorithm.
In this project, there are two part: Part 1 is the Improved AdaBoost algorithm, Part 2 is Training our own object detector. In improved AdaBoost algorithm section, two methods were used to improve the traditional AdaBoost algorithm: Weighting Parameter and limit weight expansion. Matlab was used to first emulate the traditional AdaBoost algorithm and the improved AdaBoost algorithm, then compare the accuracy and speed between these two algorithms. Part 2 introduces a method based on Haar-like features to train our own object detectors. In this thesis Orange was regarded as the target object. This process includes the preparation of positive samples and negative samples, setting the number of training stages.
The thesis was mainly carried on by using Python programming language based on window 10 operating system. OPENCV was used for the process images and training the objector detector.
In conclusion, it will become a popular field to recognizing different kinds of objects, not only faces of human beings be recognized, but also any other objects in the real world can be recognized in the future.
In this project, there are two part: Part 1 is the Improved AdaBoost algorithm, Part 2 is Training our own object detector. In improved AdaBoost algorithm section, two methods were used to improve the traditional AdaBoost algorithm: Weighting Parameter and limit weight expansion. Matlab was used to first emulate the traditional AdaBoost algorithm and the improved AdaBoost algorithm, then compare the accuracy and speed between these two algorithms. Part 2 introduces a method based on Haar-like features to train our own object detectors. In this thesis Orange was regarded as the target object. This process includes the preparation of positive samples and negative samples, setting the number of training stages.
The thesis was mainly carried on by using Python programming language based on window 10 operating system. OPENCV was used for the process images and training the objector detector.
In conclusion, it will become a popular field to recognizing different kinds of objects, not only faces of human beings be recognized, but also any other objects in the real world can be recognized in the future.