Research And Application Of Human Tracking And Fall Detection Algorithm Based On Kinect

In recent years,the aging trend of the population has become more and more serious,and the elderly have a high probability of falling because of their poor physical fitness and weak balance.The falling will have a great impact on the psychological and physical characteristics of the elderly.So it is very important to detect falls and notify contacts in time.After analyzing the traditional fall detection system,this paper proposes a detection algorithm based on human body coordinate form shift that uses the depth images and skeletal point coordinates of the elderly acquired by Kinect,and the anti-occlusion tracking algorithm of the human body are added,which further improves the tracking and recognition efficiency of the system,reduces the amount of calculation.This algorithm first addresses the problem of target lost due to occlusion in human tracking The human tracking algorithm is divided into two situations:occlusion and unocclusion.The babbitt coefficient of the image matching and the acquisition of the center point of the human skeleton are used to distinguish whether the target has been occlusion,And use different algorithms for tracking.When the target is not occluded,this paper uses the Mean shift algorithm based on the skeleton center point to track,use Kinect to obtain the skeleton point and depth image,and combine the color histogram with the human body center point.It solves the problem of inaccurate tracking when the background and human color are similar in the Mean shift algorithm.When object occlusion occurs,this paper uses extended Kalman filtering to continue to use the bone center point to predict the motion trajectory of the next frame,so that the template matching is more accurate;When the body occlusion occurs,the principle of bone nearest is adopted,and the bone features are extracted based on the bone point near the Kinect side,which makes the acquisition of coordinate points more accurate and effective.Through the human anti-occlusion target tracking algorithm,the real-time position of the human body is tracked.Based on this,the skeleton coordinate information of the human body is obtained through Kinect,and a fall detection algorithm based on human body coordinate form shift is proposed.Based on the consideration of the shape of the human body at the time of the fall,the angle analysis of the displacement of the human body shape before the fall was performed.According to the characteristics of the displacement changes of the bone joint points in the continuous frame images,the human spine point,the shoulder center point,and There are multiple bone coordinates,such as the sole of the foot,and three detection features are set,which solves the problem of inaccurate detection of a single feature.A large number of experiments were performed to select the height threshold,speed threshold,and angle threshold.By calculating the feature values in real time and comparing them with the threshold values,it was determined whether the human body had fallen.In this paper,the algorithm is applied to the system to achieve 24-hour fall detection and alarm on the human body.Once a fall is detected,the system sends email notifications and SMS reminders to related contacts.Through the test and analysis of the system,the accuracy rate of the system reached more than 94%,which verified that the system was accurate and reliable.