The Design And Implementation Of Fall Detection Based On Video

With the increasing standard of daily life in our country and the rapid development of the medical and health fields,traditional diseases have had effective treatment programs.Therefore,falling to death becomes one of the most important health problem for the elderly.In recent years,with the rapid development of computer science,pattern recognition,and modern communications,the fall detection technology gradually becomes a hot research for scholars.This paper proposes a fall detection algorithm based on video.And the algorithm consists of motion target detection,motion tracking,fusion feature extraction,and fall behavior detection.The fall detection algorithm based on video proposed in this paper is tested using Le2 i fall dataset.And the experimental results show that the accuracy of the fall detection reaches 99.30%.The main research contents of this paper are as follows: 1.Based on ViBe algorithm,this paper proposes an improved moving object detection algorithm based on Lab color space and ViBe.Since the red-green channel in Lab color space has the greatest degree of independence from shadows,this feature can be used to remove shadows.At the same time,the algorithm introduces the concept of detection result posterior probability,and the background model is updated according to the posterior probability to improve the anti-noise ability of the algorithm.Finally,the proposed algorithm is compared with ViBe,SuBSENSE and CodeBook in terms of subjective and objective evaluation.The comparison results show that the proposed algorithm has better performance than the other three algorithms in terms of anti-noise capability,real-time performance,and shadow suppression.2.In order to grasp the target motion state in real time,this paper proposes an improved target tracking algorithm based on Kalman filter.To solve the problem of low accuracy of target matching in the current target tracking algorithm,the algorithm proposed in this paper uses the comprehensive information of the color histogram of red-green channel in Lab color space and of the predicted distance to perform target matching to improve the accuracy of target matching rate.When the target object is occluded,the algorithm uses the predicted value of the Kalman filter as the final result of the tracking to effectively solve the occlusion.Finally,the single-target tracking and multi-target tracking experiments are carried out using the proposed algorithm.The experimental results show that the improved target tracking algorithm based on Kalman filter proposed in this paper has a very high degree of target matching accuracy.3.In order to accurately express the characteristics of human motion state,this paper extracts and fuses six characteristics: effective area ratio,human body aspect ratio,center rate of change,HOG feature,Hu moment invariant,and included angle between the main axis of the human body and the horizontal direction.The resulting fusion feature is an important indicator of fall detection.4.Since the accuracy of using a single frame image feature to fall detection is too low,and the time complexity of using a continuous multi-frame image feature to fall detection is too high,therefore,a three-frame fall detection method based on SVM is proposed in this paper.This algorithm extracts three frames of image fusion features with the same time interval,and uses these three frames of image fusion feature as the input value of the SVM classifier to train the classifier.Finally,the algorithm proposed in this paper is compared with the algorithm proposed by Charfi et al.on the Le2 i fall dataset.The experimental results show that the three-frame fall detection algorithm based on SVM proposed in this paper has better performance than the algorithms proposed by Charfi et al in terms of all aspects.