Real-time visual play-break detection in sport events using a context descriptor

Abstract

Abstract-This paper presents a two-stage hierarchical method for play-break detection on non-edited team sports video feed. Unlike most existing methods, this algorithm uses modern action and event recognition method thus does not rely on production cues of broadcast feeds, but instead concentrates on the content of the video. Moreover, the method does not require player tracking, can be used in real-time and can be easily adapted to different sports. In the first stage, bag-of-words event detectors are trained to recognize key events such as line changes, face-offs and preliminary play-breaks. In the second stage, the output of the detectors along with a novel feature based on the number of detected spatio-temporal interest points are used to create a context descriptor. The final classification is performed on this context descriptor. Experiments demonstrate the benefits of using this context descriptor by reducing the frame classification error by 18% when compared to the baseline method. The efficiency of the proposed method is demonstrated on a real hockey game (accuracy over 88%).

Multi-label Class-imbalanced Action Recognition in Hockey Videos via 3D Convolutional Neural Networks

“…For this, we take the basic structure, which we have described previously, and apply it to both strategies. The results of this experiment for all activities are summarized in second and third columns of Table I, respectively. 2) Comparison with Original work: The purpose of this experiment is to compare our deep learning approach for action recognition of the hockey game with the original work on the chosen dataset [15]. The results of this experiment for the three activities (as was done in [15]) are summarized in Table II.…”

“…The results of this experiment for all activities are summarized in second and third columns of Table I, respectively. 2) Comparison with Original work: The purpose of this experiment is to compare our deep learning approach for action recognition of the hockey game with the original work on the chosen dataset [15]. The results of this experiment for the three activities (as was done in [15]) are summarized in Table II. Evaluating the Use of Data Normalization.The purpose of this experiment is to evaluate our claim data normalization is important for achieving high recognition accuracy.…”

“…For action recognition of the hockey game, Carbonneau et al [15] presented a solution for play-break detection using STIP [18] detectors and SVMs. Although their system achieved a good performance, up to 90 % recognition accuracy, their analysis was limited to only three activities.…”

“…The two approaches are: (i) An ensemble of k binary networks, and (ii) A single multi-label k-output network. Also, we compare the results with a state-of-the-art existing solution [15].…”

“…We make 3D CNN our baseline and provide F 1 measure scores for a publicity available dataset [15]. We do it for 11 activities, instead of just three activities as was done in the previous work [15]. It will be useful for any researcher who is working on the same or similar problems.…”