Title page
ABSTRACT
Contents
I. Introduction 13
1.1. Background on Personalized Semantic Content Consumption in Consumer Domain 15
1.2. Background on Real-Time Video Filtering 16
1.3. Scope of Thesis 18
II. Related Works 20
III. Queueing Theory 22
3.1. Introduction of Queueing Theory 22
3.2. Basic Terminology of Queueing Theory 23
Input Source 24
Queue 25
Queue Discipline 25
Service Mechanism 26
3.3. Kendall s Notation for Classification of Queue Types 27
Examples 29
3.4. Mathematics for all Queueing Models 30
3.5. Applications of Queueing Theory 31
Traffic Flow 31
Scheduling 31
Facility Design and Employee Management 32
Some Other Examples 32
IV. Proposed Real-Time Content Filtering Framework 33
4.1. Proposed Broadcasting Content Filtering System for TV Terminals 33
4.2. Generic Filtering Processor with Visual Features 35
4.3. Applied Filtering Algorithm for Soccer Videos 39
4.4. Implementation of Real-Time Content Filtering 50
V. System Modeling 53
5.1. Analysis of Proposed Real-Time Filtering 53
5.2. Requirements of Stable Real-Time Filtering Based on D/D/1 Queue 58
5.3. Requirements of Stable Real-Time Filtering Based on D/M/1 Queue 63
VI. Experiments 72
6.1. Experimental Results on Soccer Videos 72
6.2. Filtering Requirement Analysis using D/D/1 Model 79
6.3. Filtering Requirement Analysis using D/M/1 Model 84
VII. Discussion 96
VIII. Conclusion 98
Abstract in Korean 100
References 102
Acknowledgement 107
Publications 109
Table 3-1 : Symbols meanings of Kendall s notation 28
Table 3-2 : Distribution types applied to A and B 28
Table 6-1 : Number of frames depending on frame views 74
Table 6-2 : Processing times of frame views measured in each terminal 74
Figure 1-1 : Changes of broadcasting environments 14
Figure 4-1 : System architecture of TV terminal for real-time content filtering 34
Figure 4-2 : Generic filtering processor for proposed filtering 37
Figure 4-3 : Frame view types in a soccer game : (a) global view with goal post (VGp), b) global view without goal post (VG), c) medium view (VM), and d) close-up view (VC) 41
Figure 4-4 : Filtering algorithm used for soccer videos 42
Figure 4-5 : (a) 9 sub-blocks segmented in the input frame and (b) Hue region representing the dominant color of grass in the field 45
Figure 4-6 : Accumulated edges in the region of a scoreboard 47
Figure 4-7 : Temporal view patterns of frames for shooting scenes in soccer games 49
Figure 4-8 : Screen shot to run the real-time content filtering service with a single channel of interest : (a) (1) box shows user s main broadcast and (2) box on the bottom right-hand side of the screen 51
Figure 5-1 : Simplified model of content filtering with single input 54
Figure 5-2 : Queue model of content filtering for multiple channels 56
Figure 5-3 : The flow of sampled video frames 57
Figure 5-4 : Queueing process for successive frames based on D/D/1 model 60
Figure 5-5 : Queueing process for successive frames based on D/M/1 model 65
Figure 6-1 : Performance of the proposed view decision 77
Figure 6-2 : Examples of frames extracted in each step of the filtering algorithm 78
Figure 6-3 : Variation of filtering performance according to sampling rate 80
Figure 6-4 : The number of input channels enables the real-time filtering system to satisfy the filtering requirements in (a) Terminal 1, (b) Terminal 2, and (c) Terminal 3. ① and ① lines indicate the 82
Figure 6-5 : Number of available channels according to variation of frame sampling rate in (a) P₁, (b) P₂, and (c) P₃ 87
Figure 6-6 : Regions including requirements that meet the two criteria of the proposed D/M/1 model on soccer videos in (a) P₁, (b) P₂, and (c) P₃ 91
Figure 6-7 : Buffer lengths expected and used in the proposed D/M/1 model in (a) P₁, (b) P₂, and (c) P₃ 95