Static Gesture Recognition Approach Based On DSSD

With the development of technology and the improvement of people's living standards,people are not satisfied with the manual operation of things and machines around them,but hope to achieve human-computer interaction in a simpler and faster way.In addition,due to the progress of society and the need of humanistic care,barrier-free communication between deaf-mutes and normal people is also an urgent matter to be solved,and gesture recognition is also a key part in sign language translation system.For these reasons,gesture recognition technology came into being.Traditional gesture recognition methods have the problem that accuracy and real-time can not be taken into account at the same time,and are greatly affected by the environment.In addition,these methods are not good at detecting small gestures.To solve these problems,this paper proposes a static gesture recognition method based on DSSD.The main works are as follows:Using ordinary cameras to collect pictures in different environments,and using data augmentation to expand datasets.On the basis of DSSD network model,some improvements are made: K-means algorithm and elbow method are used to re-select the aspect ratio of the priori box for the self-made dataset and common gesture datasets,so as to improve the detection accuracy;The channel fusion method is improved,the concat operation is used instead of the element product to improve the detection speed;Using migration learning,the effects of different basic networks AlexNet,VGG16 and ResNet101 on DSSD network model are compared,and the problem of low detection accuracy caused by small amount of data is solved.Subsequently,experiments verify the effect of migration learning on small data sets and select ResNet101 as the basic network of DSSD.Experiments also show that the aspect ratio selected by K-means algorithm and elbow method can indeed improve the detection accuracy and the improved channel fusion method can indeed improve the detection speed.In addition,by comparing with other gesture recognition methods,it is proved that our method has higher recognition accuracy than the gesture recognition methods based on Faster R-CNN,YOLO and SSD networks.In order to verify the stability of our method,we test it in different environments.The results show that the method is not obviously disturbed by the environment,and can complete the detection tasks in various environments.Finally,the comparative experiment proves that our method has good detection ability for small gesture and can realize gesture recognition over a long distance.Finally,a gesture unlocking system based on DSSD is designed,which can be unlocked using single gesture password or multiple gesture passwords.Experiments show the interface and performance of the system.And the results show that the system has the advantages of simple operation,high recognition accuracy,fast recognition speed and strong anti-interference ability.