16th IEEE International Conference on Advanced Video and Signal-based Surveillance (AVSS)

16th IEEE International Conference on Advanced Video and Signal-based Surveillance (AVSS)

Carlos Caetano, PhD student of computer science and member of Smart Sense, was at the international AVSS conference held in Taiwan to present the excellent results of his research on 3D action recognition using computer vision and machine learning techniques.

AVSS is the leading international conference in the field of video surveillance and signal processing. The purpose of this meeting is to provide an interdisciplinary forum that brings together experts from academia, industry and government to push the boundaries of theories and methods in this area.

Summary of the article presented:

Due to the availability of data sets that provide pose information, 3D human action recognition has drawn the attention of the scientific computer vision community.

Many works have focused on coding this information as pose image representations based on the spatial structure of the body’s joints.

To further improve these representations, SkeleMotion, a new pose image representation used as input for Convolutional Neural Networks (CNNs), was introduced. The proposed approach has the ability to encode temporal dynamics by explicitly computing the velocity and direction values ​​of the human skeleton joints.

 

 

Experimental results demonstrated the efficacy of the proposed representation in 3D action recognition, surpassing the state of the art in the NTU RGB + D 120 dataset.

http://avss2019.org/