【摘要】 为了辅助放射技师减少摆位错误，提出了一套完整的全范围头面部DR(digital radiography)智能辅助摆位系统.该系统以嵌入式计算平台为基础，利用RGB-D相机捕获的图像准确估计患者头部姿态，并针对不同的拍摄体位需求提供精确的摆位指导.采用六维向量描述头部姿态，并开发了一个深度神经网络Effi6DNet,其能够实时从RGB图像中估计全范围的头部姿态；同时，通过三维可塑模型(3DMM)和3D头部重建技术优化头部姿态估计，达到更高的准确度.该系统结合这2种技术，并融入具体摆位要求，实现了在嵌入式平台上实时和高精度的辅助摆位功能.在多个数据集上的验证结果表明，该系统能够实时和准确识别头部姿态，提供合理的摆位指导.该系统为提高头面部DR的效率和准确性提供了一种新的智能化解决方案.更多还原

【Abstract】 To assist radiographers in minimizing positioning errors, a comprehensive intelligent positioning assistance system for full-range cranium and facial bones digital radiography(DR) is introduced. Based on an embedded computing platform equipped with an RGB-depth(RGB-D) camera, this system accurately estimates a patient’s head pose, providing precise positioning recommendations for various imaging requirements. Employing a 6D vector to represent head pose, a deep neural network, Effi6DNet, is constructed to achieve real-time estimation of full-range head pose from RGB images. Furthermore, head pose optimization with enhanced accuracy is achieved through the application of 3D morphable model(3DMM) and 3D head reconstruction techniques. The integration of these two techniques, along with specific positioning requirements, facilitates real-time, high-precision positioning assistance on an embedded computing platform. Validation on multiple datasets demonstrates that the system can promptly and accurately identify head poses and offer appropriate positioning suggestions, thereby presenting an innovative intelligent solution to improve the efficiency and accuracy of the cranium and facial bones DR.更多还原