Research On Posture Recognition Methods In CT Detection

Computed tomography(CT)scanning is a commonly used modem medical scanning method.This method is convenient,rapid and safe to obtain CT images,and plays a significant role in diagnosis of disease,especially for saving time to rescue emergency patients.At present,in the CT scanning the doctor must first guide the patient in a specific position lying on the CT bed.Then the doctor should determine whether the patient is placed correctly through the laser line projected from the CT bed,which is the key to obtain high-quality and clear CT images.In order to realize the body posture recognition automatically in CT scanning and provide an automatic operational reference system for training students,a method of automatic posture recognition for CT scanning is studied in this thesis.Because there is no open source of CT position database,a CT position database is made first.By unifying the clothes and marking the joints,the position noise of CT image is reduced and the features become more obvious.The process of posture recognition including two parts:joint detection and pose estimation.In order to reduce computation and improve the speed of the algorithm,14-dimensional HOG features and a flexible mixture-of-parts model are used to detect the joint points of the human body parts.Then the datum line is extracted according to the position of joints.Finally,three factors,including the prone position,hands position and the relative position between the datum line and the laser positioning line,are used to judge whether the body position is correct.The experiment shows that the proposed method can accurately detect the human joints with a mean accuracy rate up to 92.1%.On this basis,body postures can be judged accurately,which meets the requirements of accuracy in CT detection.