当前位置:网站首页>Deep learning | palmar: adaptive multi resident activity recognition in point cloud technology

Deep learning | palmar: adaptive multi resident activity recognition in point cloud technology

2021-06-23 22:04:51 Author: snail

【 Paper title 】PALMAR: Towards Adaptive Multi-inhabitant Activity Recognition in Point-Cloud Technology

【 The author team 】Mohammad Arif Ul Alam, Md Mahmudur Rahman, Jared Q Widberg

【 Time of publication 】2021/06/22

【 machine structure 】 Department of computer science, University of Massachusetts Lowell

【 Thesis link 】https://arxiv.org/pdf/2106.11902.pdf

 

【 Recommended reasons 】 This article is from the University of Massachusetts , Considering point cloud data technology ( Laser radar 、 Millimeter wave ) high-profile , This article developed PALMAR Model ( A multi person activity recognition system ), By using efficient signal processing and novel machine learning techniques to track individuals , To develop adaptive multiplayer tracking and HAR System .

 

With the development of deep neural network and human activity recognition based on computer vision , Point cloud data technology ( Laser radar 、 Millimeter wave ) Because of its privacy protection nature, it has been widely concerned . In view of the accuracy PCD The great future of Technology , This article developed PALMAR Model , It's a multi person activity recognition system , By using efficient signal processing and novel machine learning techniques to track individuals , To develop adaptive multiplayer tracking and HAR System . More specifically , In this paper, (i) Based on voxelized feature representation of real-time PCD Fine tuning method ,(ii) Efficient clustering (DBSCAN and BIRCH), Multi person tracking and cross ambiguity reduction based on adaptive sequential hidden Markov model (iii) Novel adaptive domain adaptation technology based on deep learning , In the presence of data scarcity and diversity ( equipment 、 Location and population diversity ) In this case, we can improve HAR The accuracy of the . This article USES the (i) Three devices (3D LiDAR and 79 GHz Millimeter wave ) from 6 Real time collection of participants PCD,(ii) A publicly available 3D LiDAR Activity data (28 participant ) and (iii) The framework and system of this paper are experimented to evaluate an embedded hardware prototype system , In many people (96%) The scene offers promising HAR performance , Compared to the most advanced framework , Multiplayer tracking performance improved 63%, Without losing significant system performance in edge computing devices .

 

The architecture of this model is shown in the figure below , It mainly includes four parts , Data collection 、 Data processing 、 Multi person tracking and deep learning section .

chart 1 System architecture diagram

版权声明
本文为[Author: snail]所创,转载请带上原文链接,感谢
https://chowdera.com/2021/06/20210623220142734t.html