SPATIO-TEMPORAL LATENT FEATURES FOR SKELETON-BASED HUMAN ACTION RECOGNITION USING GCN+SOFTMAX CLASSIFIER
Authors
Zakhriddin Mominov, Avazjon Rakhimovich Marakhimov, Kabul Kadirbergenovich Khudaybergenov
Share
Annotation
Human action recognition through skeletal analysis represents a fundamental challenge with significant implications for real-world applications. Contemporary approaches frequently depend on singular skeletal sequence representations, potentially limiting their capacity to comprehensively encode the multifaceted characteristics inherent in human actions. This work introduces LFHAR (Latent Features for Human Action Recognition), an innovative architectural framework that leverages diverse spatio-temporal latent encodings to enhance action feature extraction. The proposed representations model the temporal progression of skeletal configurations while incorporating both joint-level and limb-level motion patterns. The methodology employs a graph-based transformation for individual skeletal frames within temporal sequences, subsequently organizing the extracted graph features into spatio-temporal matrices. Experiments on benchmark datasets validates the robustness and invariance properties of the LFHAR framework. The approach achieves notable performance gains, with accuracy improvements of 2.7% and 2.1% on the NTU-RGB+D 60 and NTU-RGB+D 120 datasets, respectively, substantiating its effectiveness in advancing skeleton-based action recognition.
Keywords
Authors
Zakhriddin Mominov, Avazjon Rakhimovich Marakhimov, Kabul Kadirbergenovich Khudaybergenov
Share
References:
Sun Z., Ke Q., Rahmani H., Bennamoun M., Wang G., Liu J. Human action recognition from various data modalities: A review IEEE Trans. Pattern Anal. Mach. Intell. (2022)
Ahmad T., Jin L., Zhang X., Lai S., Tang G., Lin L.. Graph convolutional neural network for human action recognition: A comprehensive survey. IEEE Trans. Artif. Intell., 2 (2) (2021), pp. 128-145
Cheng K., Zhang Y., He X., Chen W., Cheng J., Lu H., Skeleton-based action recognition with shift graph convolutional network, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 183–192.
Xin W., Liu Y., Liu R., Miao Q., Shi C., Pun C.-M. Auto-learning-gcn: An ingenious framework for skeleton-based action recognition. Chinese Conference on Pattern Recognition and Computer Vision, PRCV, Springer (2023), pp. 29-42
Liu R., Liu Y., Wu M., Xin W., Miao Q., Liu X., Li L. SG-CLR: Semantic representation-guided contrastive learning for self-supervised skeleton-based action recognition. Pattern Recognit., 162 (2025), Article 111377
Marakhimov, A.R., Khudaybergenov, K.K. Softmax Regression with Multi-Connected Weights. Computers, 2025, accepted.
Aouaidjia K., Zhang C. and Pitas I., Spatio-temporal invariant descriptors for skeleton-based human action recognition, Inf Sci (NY), 700, 121832, doi: 10.1016/j.ins.2024.121832 (2025)
Other articles of the issue
