神经心理量表理解力检测的人体姿态特征识别方法

神经心理测试可以对各认知域受损严重程度做出客观评价，是检测疾病进展、评估药物疗效的有效手段。其中理解力测试部分通过判断受试者是否根据指令要求作出相应动作实现，是老年人认知功能障碍评估的重要部分，有利于痴呆的早预防早干预。文章提出了一套神经心理测试中理解力检测的人体姿态估计视频分析方法，基于Openpose深度卷积网络提取人体关键点坐标，随后基于图像形态学处理技术和Faster R-CNN等技术提出了纸张、牙刷等目标物体关键点二维坐标提取方法，并以量表中动作要求建立人体姿态估计数学模型。通过实验对神经心理测试的6个动作进行识别，结果表明，所提姿态估计数学模型和交互动作识别方法能够有效检测人体姿态动作指令及人与纸张的交互指令。

Human posture feature recognition method for neuropsychological comprehension test

Neuropsychological test can objectively evaluate the severity of cognitive impairment. It is an effective means to detect disease progression and evaluate drug efficacy. Comprehension test is an important part of cognitive impairment assessment for the elderly. The assessment is performed by judging whether the subjects make accurate actions according to the instructions, which is conducive to the early prevention and early intervention of dementia. This study proposed a video analysis method of human posture estimation for comprehension detection in neuropsychological testing. The coordinates of key points of human body were first extracted based on OpenPose. Then, based on the image morphology processing technology and Fast R-CNN, a two-dimensional coordinate extraction method was proposed for the key points of the specified target objects, such as paper and toothbrush. Also, the mathematical model of human posture estimation was established. Six actions of neuropsychological test were tested to verify the effectiveness of the proposed method. The results show that the proposed mathematical model of posture estimation and interactive action recognition method can effectively detect human posture action commands and interactive instructions.