“Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter
This paper proposes an action recognition algorithm based on the capsule network and Kalman filter called “Reading Pictures Instead of Looking” (RPIL). This method resolves the convolutional neural network’s over sensitivity to rotation and scaling and increases the interpretability of the model as...
Main Authors: | , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2021-03-01
|
Series: | Sensors |
Subjects: | |
Online Access: | https://www.mdpi.com/1424-8220/21/6/2217 |
_version_ | 1797540459359240192 |
---|---|
author | Botong Zhao Yanjie Wang Keke Su Hong Ren Haichao Sun |
author_facet | Botong Zhao Yanjie Wang Keke Su Hong Ren Haichao Sun |
author_sort | Botong Zhao |
collection | DOAJ |
description | This paper proposes an action recognition algorithm based on the capsule network and Kalman filter called “Reading Pictures Instead of Looking” (RPIL). This method resolves the convolutional neural network’s over sensitivity to rotation and scaling and increases the interpretability of the model as per the spatial coordinates in graphics. The capsule network is first used to obtain the components of the target human body. The detected parts and their attribute parameters (e.g., spatial coordinates, color) are then analyzed by Bert. A Kalman filter analyzes the predicted capsules and filters out any misinformation to prevent the action recognition results from being affected by incorrectly predicted capsules. The parameters between neuron layers are evaluated, then the structure is pruned into a dendritic network to enhance the computational efficiency of the algorithm. This minimizes the dependence of in-depth learning on the random features extracted by the CNN without sacrificing the model’s accuracy. The association between hidden layers of the neural network is also explained. With a 90% observation rate, the OAD dataset test precision is 83.3%, the ChaLearn Gesture dataset test precision is 72.2%, and the G3D dataset test precision is 86.5%. The RPILNet also satisfies real-time operation requirements (>30 fps). |
first_indexed | 2024-03-10T13:00:29Z |
format | Article |
id | doaj.art-eb2e0bfdf6b14b7f9fd9e058e9c2b09f |
institution | Directory Open Access Journal |
issn | 1424-8220 |
language | English |
last_indexed | 2024-03-10T13:00:29Z |
publishDate | 2021-03-01 |
publisher | MDPI AG |
record_format | Article |
series | Sensors |
spelling | doaj.art-eb2e0bfdf6b14b7f9fd9e058e9c2b09f2023-11-21T11:34:00ZengMDPI AGSensors1424-82202021-03-01216221710.3390/s21062217“Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman FilterBotong Zhao0Yanjie Wang1Keke Su2Hong Ren3Haichao Sun4Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaChangchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaChangchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaChangchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaChangchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, ChinaThis paper proposes an action recognition algorithm based on the capsule network and Kalman filter called “Reading Pictures Instead of Looking” (RPIL). This method resolves the convolutional neural network’s over sensitivity to rotation and scaling and increases the interpretability of the model as per the spatial coordinates in graphics. The capsule network is first used to obtain the components of the target human body. The detected parts and their attribute parameters (e.g., spatial coordinates, color) are then analyzed by Bert. A Kalman filter analyzes the predicted capsules and filters out any misinformation to prevent the action recognition results from being affected by incorrectly predicted capsules. The parameters between neuron layers are evaluated, then the structure is pruned into a dendritic network to enhance the computational efficiency of the algorithm. This minimizes the dependence of in-depth learning on the random features extracted by the CNN without sacrificing the model’s accuracy. The association between hidden layers of the neural network is also explained. With a 90% observation rate, the OAD dataset test precision is 83.3%, the ChaLearn Gesture dataset test precision is 72.2%, and the G3D dataset test precision is 86.5%. The RPILNet also satisfies real-time operation requirements (>30 fps).https://www.mdpi.com/1424-8220/21/6/2217human posture estimationcapsule network6D object pose estimationKalman filter |
spellingShingle | Botong Zhao Yanjie Wang Keke Su Hong Ren Haichao Sun “Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter Sensors human posture estimation capsule network 6D object pose estimation Kalman filter |
title | “Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter |
title_full | “Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter |
title_fullStr | “Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter |
title_full_unstemmed | “Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter |
title_short | “Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter |
title_sort | reading pictures instead of looking rgb d image based action recognition via capsule network and kalman filter |
topic | human posture estimation capsule network 6D object pose estimation Kalman filter |
url | https://www.mdpi.com/1424-8220/21/6/2217 |
work_keys_str_mv | AT botongzhao readingpicturesinsteadoflookingrgbdimagebasedactionrecognitionviacapsulenetworkandkalmanfilter AT yanjiewang readingpicturesinsteadoflookingrgbdimagebasedactionrecognitionviacapsulenetworkandkalmanfilter AT kekesu readingpicturesinsteadoflookingrgbdimagebasedactionrecognitionviacapsulenetworkandkalmanfilter AT hongren readingpicturesinsteadoflookingrgbdimagebasedactionrecognitionviacapsulenetworkandkalmanfilter AT haichaosun readingpicturesinsteadoflookingrgbdimagebasedactionrecognitionviacapsulenetworkandkalmanfilter |