I am a Ph.D. student in the Computer Science Department at Universidade Federal de Minas Gerais (UFMG), Brazil. I am under the supervision of Erickson R. Nascimento and Mario F. M. Campos, both from the same department to which I belong. I received my MSc degree in Computer Science from Universidade Federal de Minas Gerais (UFMG) in March of 2017 and my BSc degree in Computer Science from Pontifícia Universidade Católica de Minas Gerais (PUC-Minas) in December of 2013.
I am interested in processing the information produced by wearable cameras by using the Computer Vision techniques. I am recently working on fast-forwarding egocentric videos as far as the semantic information is concerned. I have also worked on Face Recognition with respect to authentication and security.
Dissertation Title: Semantic Hyperlapse for Egocentric Videos
Authors: Washington L. S. Ramos; Mario F. M. Campos; Erickson R. Nascimento
Abstract: The emergence of low-cost personal mobiles devices and wearable cameras and, the increasing storage capacity of video-sharing websites have pushed forward a growing interest towards first-person videos. Wearable cameras, in particular, can operate for hours without the need for continuous handling. That leads these videos to be generally long-running streams with unedited content, which makes them boring and visually unpalatable since the natural body movements cause the videos to be jerky and even nauseate. Hyperlapse algorithms aim to downsize long and monotonous videos into short fast-forward watchable videos with no abrupt transitions between the frames. However, an important aspect of such videos is that some parts of them may be more important than others, so they should have their proper attention. In this work, we propose a novel methodology capable of summarizing and stabilizing egocentric videos by extracting and analyzing the semantic information in the frames. This work also describes a dataset collection with several labeled videos and introduces a new smoothness evaluation metric for egocentric videos. Several experiments are conducted to show the superiority of our approach over the state-of-the-art hyperlapse algorithms as far as semantic information is concerned. According to the results obtained, our method is on average 10.67 percentage points higher than the best method in relation to the maximum amount of semantics that can be obtained, given the required speed-up.
Authors: Michel M. Silva; Washington L. S. Ramos; Joao P. K. Ferreira; Mario F. M. Campos; Erickson R. Nascimento
Abstract: The emergence of low-cost personal mobiles devices and wearable cameras and the increasing storage capacity of video-sharing websites have pushed forward a growing interest towards first-person videos. Since most of the recorded videos compose long-running streams with unedited content, they are tedious and unpleasant to watch. The fast-forward state-of-the-art methods are facing challenges of balancing the smoothness of the video and the emphasis in the relevant frames given a speed-up rate. In this work, we present a methodology capable of summarizing and stabilizing egocentric videos by extracting the semantic information from the frames. This paper also describes a dataset collection with several semantically labeled videos and introduces a new smoothness evaluation metric for egocentric videos that is used to test our method.
Authors: Washington L. S. Ramos; Michel M. Silva; Mario F. M. Campos; Erickson R. Nascimento
Abstract: The popularity of first person videos has been increased in social media. Thanks to the low operational cost and large storage capacity of the cameras and wearable devices, people are recording many hours of daily activities, sport actions and home videos. These videos, named Egocentric videos, are long-running videos with unedited content, which make them boring and difficult to watch. A central challenge is to make egocentric videos watchable. The natural motion of the recorder’s body in a fast-forward mode becomes nauseate. In this work we propose a novel methodology to compose the new fast-forward video by selecting frames based in semantic information extracted from images. The experiments show that our approach outperforms the state-of-the-art in as far as semantic information is concerned and is also capable of producing pleasant video to be watched.
In this thesis project, we keep our research in Semantic Hyperlapse. We focus on the user's preferences to highlight the important parts of the video. Social networks are great information providers about the profile of the users. We use those information to create Semantic Hyperlapse videos, which are smooth fast-forward first-person (egocentric) videos with their relevant parts exhibited in a lower speed-up rate.
Mar. 2017 - Present
Tons of data are generated from wearable cameras and other mobile devices, but not all the information that can be drawn from these data is equally relevant. In the dissertation project, we deal with the acceleration of first-person (egocentric) videos trying to keep the smoothness and highlight the important parts without affecting the experience of the video continuity.
Aug. 2015 - Mar. 2017
During the last half of my undergraduate, I joined the face recognition research group led by Prof. Alexei Manso Correa Machado. We were challenged by the lack of face samples (a recurring problem in authentication tasks) in the algorithms training stage, which significantly decreases the recognition rate. Principal Component Analysis (PCA), Multiclass Linear Discriminant Analysis (LDA) and Two-Class Linear Discriminant Analysis (2C-LDA) were the main techniques used for us to achieve our goal.
Feb. 2010 - Dec. 2013