Development of a vision system for building concrete structures classification

Abstract

With the advancements of technology to improve the living standards of life and improving the efficiency of everyday tasks, there is an increase in robot applications. It is highly desirable to make robots intelligent so that the robots can function independently and adapt their operations to changing circumstances in an environment. With the increase of robot applications, object classification has also shown a tremendous increase in the field of computer vision. Classification identifies objects by classifying them into one of the finite sets of classes. By integrating a robot with machine learning helps to improve the efficiency and accuracy of the tasks that we are doing. Inception V3 network is an image classification model that has proved to attain an accuracy of more than 78.1% on the ImageNet dataset and the robot that the project is using is in collaboration with Control Systems Laboratory (EEE). At least 200 images for the train dataset and at least 40 images for the test dataset are used for the project. This report highlights the work and research done throughout the final year project. The objective of this project is to use Inception network to train the neural network to classify basic concrete structures classes such as stairs and pillars. This involves the collection and training of the dataset to classify the concrete structures. Furthermore, the result obtained during the experiment is explained.