Abstract:Recently, goalmouth detection has become the basis of highlight summarization and a hot topic in high-level semantic concept detection, as the appearance of goalmouth indicates a high likelihood of exciting action in field-ball game video. In addition, machine learning for object detection is a promising research frontier. A robust method is proposed to detect the presence of goalmouth frame in field-ball game video based on fuzzy decision trees. Balance process was added in training procedure. The experimental results show that our algorithm (F-measure>95%) can improve the classification performance in comparison with the threshold-based algorithm and the decision tree-based algorithm. Fuzzy rules can also be easily deduced from the constructed tree to interpret the classification model.