eduzhai > Applied Sciences > Engineering >

Real-time CNN-based Segmentation Architecture for Ball Detection in a Single View Setup

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: This paper considers the task of detecting the ball from a single viewpointin the challenging but common case where the ball interacts frequently withplayers while being poorly contrasted with respect to the background. Wepropose a novel approach by formulating the problem as a segmentation tasksolved by an efficient CNN architecture. To take advantage of the balldynamics, the network is fed with a pair of consecutive images. Our inferencemodel can run in real time without the delay induced by a temporal analysis. Wealso show that test-time data augmentation allows for a significant increasethe detection accuracy. As an additional contribution, we publicly release thedataset on which this work is based.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...