Scholarlink

In the field of gait analysis using human skeleton detection to evaluate the rehabilitation process of patients with musculoskeletal disorders after brain surgeries, MediaPipe, as a tool, has gained popularity among developers for its efficiency, real-time performance, open-source nature, and cross-platform compatibility. However, its accuracy diminishes when dealing with complex, real-world data, such as videos with intricate backgrounds. This paper analyzes the internal detection mechanisms of MediaPipe in human pose estimation and proposes an image-processing-based method to maximize its strengths by generating human-body-contour-based region of interest (ROI) from existing information to guide skeleton detection in subsequent frames. This approach mitigates the impact of complex dynamic backgrounds, showing initial improvements in detection accuracy. Additionally, the paper examines other environment-sensitive issues affecting MediaPipe’s performance, such as target brightness and scale, and provides insights into potential enhancements in these areas.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimized Human Skeleton Detection in Complex Background Videos Using MediaPipe: A Progressive Image Enhancement Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Taotao Chen,
  • Jan Kohout,
  • Karel Štícha,
  • Jan Mareš

摘要

In the field of gait analysis using human skeleton detection to evaluate the rehabilitation process of patients with musculoskeletal disorders after brain surgeries, MediaPipe, as a tool, has gained popularity among developers for its efficiency, real-time performance, open-source nature, and cross-platform compatibility. However, its accuracy diminishes when dealing with complex, real-world data, such as videos with intricate backgrounds. This paper analyzes the internal detection mechanisms of MediaPipe in human pose estimation and proposes an image-processing-based method to maximize its strengths by generating human-body-contour-based region of interest (ROI) from existing information to guide skeleton detection in subsequent frames. This approach mitigates the impact of complex dynamic backgrounds, showing initial improvements in detection accuracy. Additionally, the paper examines other environment-sensitive issues affecting MediaPipe’s performance, such as target brightness and scale, and provides insights into potential enhancements in these areas.