<p>Human Action Recognition (HAR) is an important research area in the development of surveillance application. It plays a vital role in computer vision. This paper proposes a novel descriptor called Multi-resolution Hexagonal Block-based Local Binary Pattern- Three Orthogonal Planes (MHBLBP-TOP) using multi-level wavelet decomposition to recognize human actions. The novel descriptor extracts discriminative features from multi-level wavelet decomposition of approximate and diagonal coefficients using regular and irregular hexagonal LBP on three orthogonal planes. The appearance and scale of human subjects in video sequences vary significantly with changes in viewpoint, which remains a major challenge in HAR. To overcome this limitation, multi-level coefficients enable effective recognition of multi-scale objects and facilitate the classification of various human actions in challenging multi-resolution scenarios. Additionally, the proposed descriptor exhibits robustness to grayscale illumination variations, helping recognize human actions under varying lighting conditions. Therefore, the proposed descriptor demonstrates robustness with respect to multi-resolution analysis and illumination variations. For classification, this study employs widely used methods, including Support Vector Machine (SVM), an ensemble (Random Subspace K-Nearest Neighbor) classifier, and a two-layer Feed Forward Neural Network (FFNN) for multi-class classification. The descriptor was tested on a variety of human actions and was found to be suitable for human action recognition. Its performance was evaluated across five datasets: KTH, Weizmann, UCF-11, IXMAS, and a Synthetic dataset captured using a Nikon D3400 DSLR camera. Experimental results indicate that the proposed descriptor achieves high recognition accuracy, particularly on the synthetic videos.</p>