Hand Shape and Gesture Recognition using Multiscale Template Matching, Background Subtraction and Binary Image Analysis

TL;DR

The paper tackles hand shape and gesture recognition under controlled conditions with a non-deep-learning pipeline. It leverages multiscale template matching with normalized cross-correlation $R(u,v,s)$, background subtraction, and centroid/bounding-box features to classify four templates (rock, paper, scissors, thumbs up) while maintaining a threshold-based abstention for uncertain cases. The approach achieves robust performance against translational and proximity variations, but shows sensitivity to lighting and distant hand details, indicating room for improvement with more templates and dynamic-background handling. Overall, this work demonstrates a fast, resource-efficient alternative for real-time hand-gesture tasks in data-constrained scenarios, serving as a practical baseline in HCI contexts.

Abstract

This paper presents a hand shape classification approach employing multiscale template matching. The integration of background subtraction is utilized to derive a binary image of the hand object, enabling the extraction of key features such as centroid and bounding box. The methodology, while simple, demonstrates effectiveness in basic hand shape classification tasks, laying the foundation for potential applications in straightforward human-computer interaction scenarios. Experimental results highlight the system's capability in controlled environments.

Figures (6)

• Figure 1: Templates used for matching.
• Figure 2: An example of background subtracted and thresholded image.
• Figure 3: Result of applying all the above Methods
• Figure 4: Scissors classified at two different lighting conditions
• Figure 5: Paper classified at two different positions