Autors: Hristov, P. A., Avresky D., Boumbarov, O. L.
Title: Human-Object Interaction Detection: 1D Convolutional Neural Network Approach Using Skeleton Data
Keywords: 1DCNN, skeleton, interaction, object, video

Abstract: Human-object interaction detection is a somewhat recently emerged scientific topic, which is mainly due to the advent of deep learning algorithms. Most current methods are performed on single images, detecting separately humans and objects, using state -of-the-art pose detection and object detection networks. The networks ease the overall task by allowing for learning of the readily inferred features. When adding the time dimension into the equation, this task becomes more complex, as temporal features between frames have to be taken into account. The paper aims to show an approach for detecting human interactions in videos, which utilizes several different methods – YOLOv5 for object detection, CSR-DCF and Kalman Filter for object tracking, and 1D Convolutional Neural Network (1DCNN) for real-time interaction detection.

References

    Issue

    NCA 2021, pp. 1-5, 2021, United States, IEEE, DOI 10.1109/NCA53618.2021

    Copyright IEEE

    Full text of the publication

    Вид: публикация в международен форум, публикация в реферирано издание, индексирана в Scopus