Real-Time Hand Gesture Recognition Using HOG-SVM: A Computationally Efficient Approach for Assistive Technologies

This research proposes a real-time hand gesture recognition system for assistive technology, mainly for sign language translation. This system is meant to achieve maximum computational efficiency for real-time applications on resource-constrained devices. Histogram of Oriented Gradients (HOG) feature extraction and Support Vector Machine (SVM) are utilized by the proposed method for the task of gesture recognition. Since it can extract both edge-based and shape-based features of the gesture, this method can be used for the detection of dynamic and static gestures. This system relies on the pre-processed set of varying gestures presented in the Word-Level American Sign Language (WLASL) dataset, which is readily accessible, to be effective in varying circumstances of illumination, backgrounds and hand pose. Pre-processing techniques such as grayscale conversion, noise removal and frame resizing were used for feature extraction, helping in the improvement of feature extraction at a very low computation cost. HOG features that possess useful gesture boundaries and shape were extracted from the pre-processed frames. Features were then classified by SVM with the help of an RBF kernel. SVM classifier with the RBF kernel has an efficiency in recognizing the gestures at the variation of non-linear shapes with an accuracy of 97.8%. This system’s performance can be increased using the hybrid features by using HOG and Local Binary Patterns (LBP) together. Furthermore, HMMs are used to model the temporal dependencies for smooth recognition of dynamic gestures while applying data augmentation techniques like rotation, scaling, and flipping to maximize the generalization and adaptability of the model. The proposed system was compared with deep learning models such as CNNs, ResNet, and MobileNet. Although deep learning models achieved slightly higher accuracy, they required significantly higher computational resources. The proposed HOG-SVM model achieved a better trade-off between accuracy and computational efficiency for real-time applications. This study demonstrates the model’s potential to integrate into sign language translation systems, smart assistive devices, and gesture-controlled interfaces as a practical solution for improving human-computer interaction.