Application Of Abnormal Posture Detection In Remote Monitoring

In the scene of remote monitoring,due to the complexity of the scene of remote monitoring,the traditional methods of abnormal posture detection are difficult to acquire satisfactory results,and the method of abnormal posture detection of human body by video monitoring has the advantages of simple equipment,low cost and convenient maintenance.Therefore,it is necessary to study the abnormal posture detection technology for bedridden patients and braking patients in the scene of remote monitoring.(1)To solve the problem that it is easy to cause injury when bedridden patients with abnormal posture are performed remote operations such as back lifting and turning over,a method of automatic recognition for abnormal posture of bedridden patients in the scene of remote monitoring is proposed.Firstly,the improved Mask R-CNN + MROI Align algorithm is used to extract object contour,and the RPN network parameters are improved to make the candidate region closer to the real boundary of the object in the scene,the ROI Align is replaced by MROI Align to make full use of the characteristics of different layers and improve the accuracy of object contour extraction.Secondly,according to the scene information and the positional relationship of the object contours,the appropriate features are selected,and a feature weighted difference discriminant method is constructed based on the centroid distance and the regional overlap rate to distinguish the people in the scene.Finally,according to the results of personnel differentiation,a contour comparison algorithm based on corresponding points is designed to recognize the abnormal posture of bedridden patients in the scene of remote monitoring.(2)Aiming at the problem of missing detection caused by the occlusion between the bed and the braking patient,the detection algorithm is proposed for braking patients based on P-Mask R-CNN + JFPN.According to the shape information of the objects,the abnormal posture is detected,the process of detection is as follows: firstly,in order to improve the detection precision of abnormal posture of braking patients,the RPN network based on the aspect ratio of real object is constructed,and JFPN module is added in the process of mask extraction,the separate detection of foreground and background objects is realized by Mask R-CNN + JFPN algorithm;then,P-Mask R-CNN + JFPN algorithm is used to fuse foreground and background objects;finally,according to the characteristics of the projective images,a similar determination algorithm of projective images based on the combination of endpoints and pixels is designed,by analyzing between the object endpoints and the similarity of the projective images,the abnormal posture of braking patients is detected.Compared with the method of abnormal posture detection based on Mask R-CNN algorithm and other methods of abnormal posture detection based on features,the detection effect of the method of automatic recognition for abnormal posture of bedridden patients in scene of remote monitoring is better,and the effect of detection is the best when the monitoring camera is fixed above the bed tail.Compared with the detection algorithm based on shadow window,the time complexity of the two algorithms is equivalent,but the accuracy of the algorithm based on P-Mask R-CNN + JFPN is improved by 7.59%.Experimental results demonstrate that the proposed method is suitable for the above scenes and achieves better results.