Research On Gait Recognition Algorithm Based On Time Series And Deep Learning

As an emerging field of biometrics recognition,gait recognition has received more and more attention from researchers in recent years.Compared with other biometric recognition technologies,gait recognition has the advantage that it is not easy to disguise and can be obtained at a long distance without contact.With the popularization of monitoring,massive amounts of data need to be analyzed and processed.It is obviously unrealistic to rely solely on manpower to process and analyze.Therefore,the study of gait recognition is particularly important.The methods of gait recognition are mainly divided into two categories.One is to treat gait as an image,compress the gait profile into an image,and convert it into image similarity matching problem.Timing information;the other is to treat gait as a video sequence,and send the gait contour sequence to LSTM or 3D-CNN network for training to extract features.But the calculation is expensive and difficult to train.In order to reduce training time,this paper extracts gait feature time series from human body contours,and studies gait recognition classification based on time series.The main work of this article includes:(1)Using the background pruning method,extract the human body contour sequence from the original video,and perform morphological closure operations on the upper half of the contour to obtain a better effect contour map;(2)By modeling the contour of the human body,estimating the position of the joint points of the human body,obtaining the corresponding gait feature time series,combining the obtained multiple gait feature time series,and using the fully convolutional neural network to classify the scale of the problem is similar to the current leading method;(3)Convert the single-view gait view to any other view by generating an adversarial network,compare the recognition accuracy under different angle data sets,and analyze the conversion of the gait view to other views except the side view feasibility.Experimental results show that the contour map extracted in this paper can obtain better recognition effect.At the same time,with time series as input,the gait recognition method based on fully convolutional neural network has achieved a recognition effect close to other deep learning methods on a small-scale data set and greatly reduced training time.