عنوان پایان‌نامه
آشکارسازی رویداد کرنر در تصاویر ویدیویی بازی فوتبال با استفاده از ردیابی توپ
- رشته تحصیلی
- مهندسی برق-مخابرات-سیستم
- مقطع تحصیلی
- کارشناسی ارشد
- محل دفاع
- کتابخانه دانشکده برق و کامپیوتر شماره ثبت: E2034;کتابخانه مرکزی -تالار اطلاع رسانی شماره ثبت: 52756
- تاریخ دفاع
- ۱۷ خرداد ۱۳۹۰
- دانشجو
- ژیلا حسین خانی
- استاد راهنما
- حمید سلطانیان زاده
- چکیده
- تحلیل معنایی تصاویر ویدئویی یک مبحث تحقیقاتی بسیار جدید و جذاب است که طی چند سال گذشته ظهور پیدا کرده و در همین مدت توجه بسیاری از محققین را به خود جلب نموده است . در واقع تحلیل معنایی ویدئو و خلاصه سازی آن دانشی است که به اراده راهکارهایی جهت جستجو و بررسی محتوا ی داده های ویدئویی مدیریت محتوای و تحلیل معنایی آنها می پردازد. معمولا خلاصه سازی بر اساس استخراج رویدادها و مفاهیم به وقوع پیوسته در ویدئو صورت می پذیرد.
- Abstract
- Video semantic analysis is a very new and interesting field of research that has emerged during the past few years and has attracted many researchers. Indeed, video semantic analysis and video summarization is the knowledge that offers solutions to search contents of video streams, manage video contents, and analyze its semantic. Summarization is usually carried out based on events and concepts occurred in the video streams. Therefore, detection of the main event is very important. In this thesis, detection of the ball passing the end-line of the soccer game yard has been studied for soccer video analysis. Considering the great popularity of soccer in comparison with other sports, its video streams have been chosen for this work. Here, by studying recent scientific articles, relevant guidelines are investigated to identify the crossing of the ball and the end-line and then an appropriate scheme is proposed to reach this objective. To this end, at first, the end-line is detected using Hough transform. Then, the ball is detected using its color, size, and shape features and tracked by the Kalman filter to obtain its trajectory. Finally, the crossing of the ball and the end-line is detected using the ball trajectory and end-line position by calculating the distance between the ball’s centroid and the end-line. Although the main goal of this thesis is to achieve high-level features from the soccer video for semantic analysis of the video, the results can be utilized to help assess team tactics, football training, and help referees make more accurate decisions on events. The proposed algorithms are tested using three different video streams. Based on the experimental results, ball detection takes 0.33 to 0.45 seconds per frame on average. The precision of this method is 74%. The ball detection algorithm yields superior results in comparison with other common methods such as template matching and circular Hough transform (CHT). It has 15% higher precision than CHT and its performance is superior to that of template matching. Also, to track the ball using the Kalman filter, around 0.15 seconds per frame is required for calculations and processing. The precision of the ball tracking step in the best case is 91%. It has 11% improvement in comparison with a tracking method that utilizes a neighborhood search window. Keywords: Segmentation, Tracking, Video Summarization, High-Level Features, Hough Transform, Kalman Filter, Soccer
