![]() The fastest serve according to the official records from the Association of Tennis Professionals is John Isner's 253 kilometers per hour at the 2016 Davis Cup. This is because badminton travels much faster than tennis, resulting in much more unclear object images in badminton videos. Table 2: Parameters used in model training ParametersĬompared with tennis tracking, it can be observed that tennis tracking outperforms badminton tracking by a noticeable margin. Table 2 shows the model training parameters, including learning rate, batch size, steps per epoch, number of epochs, etc. Any number of consecutive input frames are allowed. Note that TrackNet framework is scalable. For convenience, TrackNet that takes single input frame is named as Model I and TrackNet that takes three consecutive input frames is named as Model II. To compare the performance of TrackNet frameworks with one single input frame and three consecutive input frames, two versions of TrackNet are implemented. To speed up the training speed, all frames are resized from 1280x720 to 640x360. 70% frames are the training set and 30% frames are the test set. The dataset contains 20,844 frames and is randomly divided to the training set and test set. Both Archana's algorithm, a conventional image processing technique, and the proposed TrackNet are evaluated. The tennis dataset comes from the video of the men's singles final at the 2017 Summer Universiade. Symmetric numbers of upsampling layers and maximum pooling layers are implemented.īefore the evaluation of the badminton dataset, our previous experimental results on tennis tracking is introduced. To realize the pixel-wise prediction, upsampling is applied to recover the information loss from maximum pooling layers. The 14-24 layers refer to DeconvNet for semantic segmentation. The first 13 layers refer to the design of the first 13 layers of VGG-16 for object classification. The input of the proposed network can be some number of consecutive frames. The implementation details of TrackNet is illustrated in the following figure. The coordinates of the ball are available in the labeled dataset and the variance of the Gaussian distribution refers to the diameter of badminton images. The ground truth of the heatmap is an amplified 2D Gaussian distribution located at the center of the badminton. TrackNet is trained to generate a probability-like detection heatmap having the same resolution as input frames. TrackNet with more than one input frame can improve the moving object tracking by learning the trajectory pattern. One input frame is considered the conventional CNN network. The number of input frames is a network parameter. It takes consecutive frames to generate a heatmap indicating the position of the object. TrackNet is composed of a convolutional neural network (CNN) followed by a deconvolutional neural network (DeconvNet). Peng, “TrackNet: A Deep Learning Network for Tracking High-speed and Tiny Objects in Sport Applications”, Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining KDD ‘19 (Submitted) Dataset The precision, recall, and F1-measure of TrackNet reach 85.0%, 57.7%, and 68.7%, respectively. The network is evaluated on the video of 2018 Indonesia Open Final - TAI Tzu Ying vs CHEN YuFei. TrackNet takes images with the size of 640x360 to generate a detection heatmap from several consecutive frames to position the ball and achieve high precision even on public domain videos. The proposed heatmap-based deep learning network is trained to not only recognize the ball image from a single frame but also learn flying patterns from consecutive frames. In this work, we develop a deep learning network, called TrackNet, to track the badminton from broadcast videos in which the ball images are small, blurry, and sometimes with afterimage tracks or even invisible. ![]() Although vision-based object tracking techniques have been developed to analyze sport competition videos, it is still challenging to recognize and position a high-speed and tiny ball accurately. Ball trajectory data are one of the most fundamental and useful information in the evaluation of players' performance and analysis of game strategies.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |