Automatic Weak Learners Selection for Pattern Recognition and its application in Soccer Goal Recognition

In this paper, we propose an automatic weak learners selection approach to perform advanced weak learners. In pattern recognition, the weak learners play a critical role in order to explain distinguishable features. Our approach relies on analyzing the weak learners by means of probability density functions (PDFs). Since the PDFs are appropriate statistical tools for computer vision applications that can explain objects very well, we propose to use them as the discriminant factor. In our work, we compute a new measurement for the common surface under the PDFs curves. According to our automated measure, we decide about the weak learners, whether they are eligible or not for recognition purposes? And also how much they are appropriate for distinguishing the target objects within the images. For evaluating our approach, we recognize the soccer goals in soccer videos by means of selected weak learners. The experimental results on real-world videos show the success of our approach and its superiority in comparison to the approaches that use all weak learners.