Indoor Human Fall Detection Based On DaVinci Platform

Human action recognition based on video is a hot research area in the field of computer vision.Solutions to this problem have wide applications in domains such as intelligent surveillance,behavior recognition and human-computer interaction.Based on the current research situations,in this paper,a new vision-based action recognition method has been proposed for indoor action analysis,and we have also accomplished the software design in DaVinci platform.The proposed solution can be applied in the elderly apartment or empty-nesters in order to analyze whether someone has fell down.In this paper,the algorithm principle of human action recognition and the software design process have been elaborated.The main contributions are described as follows:1.In order to solve the problem of foreground detection of moving body,background modeling method based on HS mixture Gaussians has been proposed in this paper.We use the H and S component in HSV color space to build the background model,and we also improve the way of updating parameters and telling the foreground pixels.Shadow subtraction and morphological processing method also have been proposed to improve the detection effect.2.In order to solve the problem of human fall action modeling and classifying.For the first,feature extraction method based on motion features and morphological features is proposed in this paper.This method can make use of motion change information and morphological change information to describe the motion characteristics.In this paper,the action consists of 30 consecutive images,and we have proposed one method based on motion feature change to realize the segmentation of image sequences from video.After we get the feature vectors,the action recognition problem becomes a time series analysis problem.We use the left-right hidden Markov model which the observation variables of each hidden state subject to Gaussian distribution.3.In order to solve the problem of software realization of fall detection algorithm,we choose DaVinci platform with TMS320DM6446 processor as the hardware platform.By aid of the processor's ARM and DSP dual-core architecture,we have designed efficient programming structure which ARM-core used for task scheduling and DSP-core used for algorithm processing.We also improve the feature extraction and motion recognition design flow based on the proposed algorithm in order to improve the detection accuracy.