Object detection is an important part of the image processing system, especially for applications like Face detection, Visual search engine, counting and Aerial Image analysis. Hence the performance of object detectors plays an important role in the functioning of such systems. With the advancements in the Deep learning field, Convolutional Neural Networks (CNN) is now the state of the art in object detection and classification. In this thesis, we have investigated different object detection models and studied the chain of developments which has taken place from one model to the next, the improvements over the previous models are studied by comparing the model architecture, methods of extracting features and by the performance on different object detection and classification competitions. The thesis is divided into three parts, first, a few notable models developed before the deep learning era are studied, these models have hand-coded feature extractors. In the next part of the thesis, Convolutional Neural Networks, part of deep learning for object detection has been researched by understanding the basic Convolutional Neural Network architecture and then different ways to improve it. It is seen that the CNN object detectors use a base architecture, these architectures can be classified as basic and advanced. In the final part of the thesis, the Meta architectures, which are divided as region based and regression based models are presented in detail, with depictions of such implementations. These Meta architectures are based on the basic and advanced architectures studied in the previous sections. With a look at the hand coded features based object detectors in the beginning, this work ends with a comparison of the state-of-the-art models discussed in the Meta architecture section. In the end, the future scope of the object detectors and possible new applications are discussed. Keywords: Neural Networks, Convolutional Neural Networks, object detectors, Deep learning, regression based, region based / A Thesis submitted to the Department of Electrical and Computer Engineering in partial fulfillment of the requirements for the degree of Master of Science. / Summer Semester 2018. / July 16, 2018. / Convolutional Neural Networks, Deep learning, Neural Networks, object detectors, region based, regression based / Includes bibliographical references. / Simon Foo, Professor Directing Thesis; Bruce A. Harvey, Committee Member; Pedro Moss, Committee Member.
| Identifer | oai:union.ndltd.org:fsu.edu/oai:fsu.digital.flvc.org:fsu_647274 |
| Contributors | Panthula, Ganesh Anirudh (author), Foo, Simon Y. (professor directing thesis), Harvey, Bruce A., 1961- (committee member), Moss, Pedro L. (committee member), Florida State University (degree granting institution), FAMU-FSU College of Engineering (degree granting college), Department of Electrical and Computer Engineering (degree granting departmentdgg) |
| Publisher | Florida State University |
| Source Sets | Florida State University |
| Language | English, English |
| Detected Language | English |
| Type | Text, text, master thesis |
| Format | 1 online resource (70 pages), computer, application/pdf |
Page generated in 0.0051 seconds