Nowadays, deep learning has been extensively supported in various healthcare applications to detect diseases from the human body. It becomes so powerful when it is combined with other domains like deep learning and machine vision. The conventional methods of healthcare monitoring systems only contain smart wearables and vision-based methods that are limited to detecting only the specific issues from the human body. The proposed system gives a detailed summary and experiments on a deep learning-based non-invasive disease diagnosis approach in a smart healthcare system. The proposed system is processed into two steps: first, AX-YOLOV5 (Arbitrary Extra-Large You Only Look Once Version 5) algorithm is to detect the position of the patient in the video input, and next the AlphaPose Library is to detect 17 key points from the patient’s body, and their body movements are continuously captured by an IP camera. AX-YOLOV5 algorithm analyzes the real-time images from the video sequences to localize the position of the patient’s body. The key points are compared with the five most important key point coordinates of the human body using the rule of mining associations. These key points are used to identify the body position of a patient either lying on a bed or sitting. The temporal thresholding technique recognizes healthcare issues by how repeatedly the coordinates of the key points of the human body move in a certain period. Last, the distance of key points and the temporal threshold helps to categorize the diseases of the human organs. Moreover, the coordinates of the key points are accessed for identifying the correct disease from the human body. Based on the experimental results, the confusion matrix created by the proposed system reveals the accuracy is 99%.