Research On Motion Gesture Detection And Recognition Technology Based On Wi-Fi

With the rapid development of artificial intelligence,gesture recognition has received widespread attention as a new generation of human-computer interaction technology.Traditional wearable device-based gesture recognition technology requires users to wear a proprietary device,which cannot be used when the battery is low and easily causes user inconvenience;computer vision-based gesture recognition technology relies on high-resolution video or image signals,and it is useless in smoke and dark conditions;In radio frequency,gesture recognition based on proprietary RF devices limits the promotion of gesture recognition due to the large and expensive equipment,but gesture recognition technology based on commercial Wi-Fi signals is easy to popularize because it does not require additional equipment and lighting conditions,and has the advantages of low overhead and easy deployment.The existing gesture recognition methods have the following characteristics: First,lacking of mature gesture behavior detection extraction methods.Gesture behavior detection extraction is the basis for gesture recognition and has a large impact on the classification results.The current gesture detection and extraction algorithms are not suitable for practical environments.Second,the feature types are relatively single.Most of the existing gesture recognition algorithms only consider only time domain features or frequency domain features.Third,it is difficult to achieve a balance between training time cost and classification accuracy.Based on the above characteristics,this paper will carry out research on gesture detection extraction and recognition technology based on Wi-Fi signals.The main contents are as follows:Firstly,research on algorithms for gesture behavior detection and extraction.Using the principle of intrusion detection as a reference,the kernel density estimation method is used to estimate the probability density function in the silent state,and the initial detection threshold is obtained.In addition,the silent data monitored in the online phase is added to complete the real-time update of the detection threshold,and finally combined with the cache window method to extract gesture behavior.The algorithm was tested and analyzed in the outdoor open environment and indoor multipath environment.The results show that the algorithm can effectively improve the detection rate and complete the extraction of gesture behavior.Secondly,feature extraction for a single gesture behavior to build a feature vector matrix.Aiming at the problem of a single feature type,joint mean,quartile,autocorrelation coefficient and other time domain change features,spectrum entropy,spectrum energy and other frequency domain change features jointly describe gesture behavior.At the same time,in order to avoid the problem of low recognition rate caused by inconsistent feature ranges,all features are normalized by using the zero-mean normalization method.Thirdly,research on algorithms for gesture recognition.Based on the support vector machine classification algorithm and the particle swarm optimization algorithm,a support vector machine classification algorithm based on the improved particle swarm algorithm is studied.The test and analysis are performed in the outdoor open environment and indoor multipath environment.The results show that the algorithm can quickly converge to a globally optimal position while balancing global search and local search,and effectively improves the recognition rate of gesture behavior.