Mask segmentation and classification with enhanced grasshopper optimization of 3D hand gestures
The difficulties associated with extracting 3D hand meshes from depth image utilizing 2D convolutional neural networks. The precision of such estimations is frequently hampered by visual distortions caused by nonrigidity, complex backdrops, and shadows. This research provides a unique methodology...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2023
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/10831/1/24p%20FAWAD%20SALAM%20KHAN.pdf http://eprints.uthm.edu.my/10831/2/FAWAD%20SALAM%20KHAN%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/10831/3/FAWAD%20SALAM%20KHAN%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The difficulties associated with extracting 3D hand meshes from depth image utilizing
2D convolutional neural networks. The precision of such estimations is frequently
hampered by visual distortions caused by nonrigidity, complex backdrops, and shadows.
This research provides a unique methodology that combines the enhanced grasshopper
optimization method for feature optimization with MASK-RCNN and FCN for
segmenting and classifying 3D hand gestures to address these problems. In order to
evaluate the proposed method, a 3D gesture data set is generated. In addition, a skeleton
model for RGB hand gestures is constructed by estimating the degree of freedom (DoF)
using human kinematics. The segmentation of 3D hand gestures is computed using the
ResNet50 backbone network, and the Overlap Coefficient (OC) is employed as an
evaluation metric. On the other hand, the ResNet101 backbone network is used to
calculate the classification of 3D hand gestures. Experimental results reveal that the
proposed method achieves greater accuracy in segmenting and classifying 3D hand
gestures than existing methods. The study also emphasizes the significance of using
feature optimization approaches and developing skeletal models to estimate (DoF) in
order to improve the precision of 3D hand gesture analysis. This study provides a
promising approach for robust and precise 3D hand gesture recognition, with potential
applications in disciplines such as human-computer interaction and virtual reality. The
test results show best accuracy for 3D hand gesture classification and segmentation. On
the private dataset, the classification accuracy is 99.05 %, whereas 99.29 % on the Kinect
dataset, 99.39 % and 99.29% using SKIG and ChaLearn dataset during validation. The
OC is 88.16 % and 88.19 %, respectively which is the highest available accuracy
compared with other methods. The mAP of ChaLearn 93.26%, private 81.48%, SKIG
75.21% and Kinect 66.74%. |
|---|