Small Target Detection in Human-Robot Interaction: A Research and Application Analysis

Beijing 21 Century School, Beijing, China

This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

The advancement of computer vision and robot vision techniques has revolutionized human-robot interaction by enabling more precise detection of small targets, particularly in challenging environments. This paper presents a comprehensive analysis of remote gesture recognition systems that address the limitations of traditional visual inspection methods, such as interference from cluttered backgrounds, low-resolution inputs, and partial occlusions. A detailed investigation of state-of-the-art algorithms—including Faster R-CNN, Mask R-CNN, SSD, RetinaNet, and YOLO variants—reveals persistent challenges in feature loss, scale sensitivity, and computational inefficiency during gesture detection in complex scenarios. The research explores innovative strategies that integrate multi-scale progressive fusion methods, such as the Adaptive Feature Pyramid Network (AFPN) and adaptive spatial fusion techniques, to enhance detection performance. Enhancements to YOLO-v8 combined with a shape-IoU loss function further improve the accuracy and robustness of small gesture recognition. The approach employs advanced deep learning, sensor fusion, and edge computing techniques to refine the gesture recognition process, from data acquisition by various sensors—including monocular cameras, multi-ocular cameras, and depth sensors—to sophisticated preprocessing, segmentation, and feature extraction. Comparative experimental results indicate that high-precision models like Mask R-CNN deliver superior accuracy, while optimized lightweight frameworks such as YOLO-v8 achieve real-time performance at 30+ frames per second, making them highly suitable for dynamic applications.