A Vision-Based Target Localization Method for Robot-Assisted Sonography

This article proposes a vision-based ultrasound scanning localization method that solves the issue of human pose constraints. Specifically, by proposing an improved AlphaPose network, the neck positioning for individuals with different body types is addressed. The interference caused by other moving human subjects in the scene is resolved through an improved DeepSORT tracking algorithm. The impact of environmental noise interference is reduced through depth information processing techniques. Additionally, a robotic platform for neck ultrasound scanning is constructed. This system is capable of automatically identifying and localizing moving human necks, planning trajectories for the ultrasound probe approach, and achieving real-time interaction.