RESEARCH ON 3D RECONSTRUCTION METHOD AND APPLICATION OF FOOD IN STROKE PATIENTS BASED ON RGB-D IMAGE

RGB-D images can be used to build 3D models with high precision and efficiency. This paper introduces the method and principle of 3D modeling and the 3D reconstruction process of RGB-D images and analyzes KinectFusion and BundleFusion methods. Based on BundleFusion, this paper constructs a set of 3D reconstruction methods for stroke patients’ food and carries out 3D reconstruction experiments on watermelon, banana, orange, corn, and dish cold bitter gourd. With the actual measurement data validation, the banana’s average error rate of 3.95%; for watermelon, the average error rate is 3.89%; for grapes, the average error rate is 4.28%; for corn, apple, banana mixture composition, the average error rate is 2.61%, the average error rate of caramel treats is 3.48%, and the average error rate of pork floss bread is 2.85%. The average error rate of the hot and dry noodles is 1.85%, and the average error rate of the breakfast mix composition is 2.91%. The error rate in subsequent food volume calculation, quality estimation and calculation application is within the effective range of nutrients. It provides a technical basis and scientific basis for dietary intervention of stroke patients.