【摘要】 坐姿检测在医疗健康、自动驾驶、辅助课堂教学等领域都有较高的应用价值。实际应用中，坐姿检测技术的速度与精度都需要达到较高的水平，为此，基于双目视觉，设计一种实时坐姿检测的方法。首先利用双目摄像头实时采集用户图像，然后利用简化的OpenPose模型提取人体骨骼关键点并根据关键点坐标计算头部倾斜角，接着用改进的半全局匹配算法获取双目图像的视差，并计算相应关键点的深度信息，由骨骼关键点坐标、头部倾斜角以及关键点的深度信息作为特征，设置阈值检测坐姿，当用户采取非正确坐姿时，进行语音播报提醒。更多还原

【Abstract】 Sitting posture detection has a high application value in the fields of health care, automatic driving, auxiliary classroom teaching and so on.In practical application, the speed and precision of sitting posture detection technology need to reach a high level.Therefore, based on binocular vision, a real-time sitting posture detection method is designed.By using binocular camera real-time gathering user image first, and then OpenPose algorithm is utilized to extract human body skeleton and according to the key point coordinate calculation head tilt Angle, with improved then half global matching algorithm to obtain the parallax of binocular image and calculate the corresponding key points of depth information, by bone point coordinates, head tilt Angle and the depth of information as a characteristic of key, set the threshold detection sitting position, when users take the correct posture for voice broadcast to remind.更多还原