Browsing by Author "Kara, Yunus Emre."

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Computer vision-based human action recognition via keypoint tracking
    (Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2011., 2011.) Kara, Yunus Emre.; Akarun, Lale.
    Computer vision-based human action recognition is a highly active research area which has many application areas including security, surveillance, assisted living, and entertainment. In this thesis, a new system for computer vision-based recognition of human actions is presented. The proposed system uses videos as input. The approach is invariant of the location of the action and zoom levels, the appearance of the person, partial occlusions including self-occlusions and some viewpoint changes. It is robust against temporal length variations. Keypoints are tracked through time and the trajectories of tracked keypoints are used for interpreting the human action in the video. Then, features from videos are extracted. A group of features for describing a trajectory are proposed. Trajectories are clustered using these trajectory features. The clustered trajectories are used for describing an image sequence. Image sequence descriptors are the normalized histograms of the clusters of trajectories. At the nal stage, the proposed system uses the descriptors of the image sequences in a supervised learning approach. An application based on the proposed method has been developed and applied to various datasets. A new multi modal dataset, called WeCare, which is focused on elderly care systems is introduced. The main objective of the dataset is to detect falls of humans. For attaining this goal, some other actions that can be confused with the falling action are included in the dataset. The evaluation of the proposed approach is done using two datasets: KTH Human Action Dataset and URADL Dataset. The proposed technique performs comparable to the methods in the literature. It has 87.25 per cent accuracy on the KTH dataset, 88 per cent accuracy on the URADL dataset. It has an accuracy of 98.75 per cent on the WeCare dataset.
  • Thumbnail Image
    Item
    Crowd - labelling for continuosun - valued annotations
    (Thesis (Ph.D.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2018., 2018.) Kara, Yunus Emre.; Akarun, Lale.
    As machine learning gained immense popularity across a wide variety of domains in the last decade, it has become more important than ever to have fast and inexpensive ways to annotate vast amounts of data. With the emergence of crowdsourcing services, the research direction has gravitated toward putting ‘the wisdom of crowds’ to use. We call the process of crowdsourcing based label collection crowd-labeling. In this thesis, we focus on crowd consensus estimation of continuous-valued labels. Unfortunately, spammers and inattentive annotators pose a threat to the quality and trustworthiness of the consensus. Thus, we develop Bayesian models taking different annotator behaviors into account and introduce two crowd-labeled datasets for evaluating our models. High quality consensus estimation requires a meticulous choice of the candidate annotator and the sample in need of a new annotation. Due to time and budget limitations, it is beneficial to make this choice while collecting the annotations. To this end, we propose an active crowd-labeling approach for actively estimating consensus from continuous-valued crowd annotations. Our method is based on annotator models with unknown parameters, and Bayesian inference is employed to reach a consensus in the form of ordinal, binary, or continuous values. We introduce ranking functions for choosing the candidate annotator and sample pair for requesting an annotation. In addition, we propose a penalizing method for preventing annotator domination, investigate the explore-exploit trade-off for incorporating new annotators into the system, and study the effects of inducing a stopping criterion based on consensus quality. Experimental results on the benchmark datasets suggest that our method provides a budget and time-sensitive solution to the crowd-labeling problem. Finally, we introduce a multivariate model incorporating cross attribute correlations in multivariate annotations and present preliminary observations.