Research On Gait Recognition Strategy And Its Model

With the success of computer technology and a series of mathematical rules in dealing with the development of various practical problems,deep learning and artificial intelligence have become the focus of attention and research in various countries,and their applications have now spread to all areas of people's lives.However,based on the existing deep learning technology,the classification recognition in the field of processing image and video changes with the resolution of image and video,and the timeliness of recognition will change greatly.In order to reduce training learning and identify classification costs and increase application flexibility,the use of sensor-based data sets has unparalleled advantages in real production and life.Although the use of sensors for motion recognition has a natural advantage: the compactness of the data structure and data set makes motion state recognition based on sensor data sets very convenient in practical human life applications.Regardless of the type of sensor used,the recognition of the motion state is affected by the noise generated by the physical system and the corresponding users,the use environment,and the system itself.The noise becomes an important component of the sensor generated data,and many excellent The authors of the paper propose a series of algorithms and models to solve the accuracy of data recognition.These papers do their best to remove the influence of noise in the process of motion state recognition.However,it is almost impossible to accurately distinguish between noise data and non-noise data by algorithms in actual scenes.Moreover,the traditional single model cannot identify noise.The first important processing principle of our proposed new model is to treat noise data and non-noise data as a whole.This completely avoids the complexity of distinguishing between noisy and non-noisy data,and is suitable for the recognition mode of the motion state recognition mode of different sensor noises and the heterogeneous user motion state(different age groups and diverse use environments).The second important feature of the model is to use the local clustering algorithm in a non-global unified way to approximate the optimal solution of local features,which has incomparable advantages for the recognition of motion states.The third important feature of the model is the conversion of single-label tasks into multi-label tasks,retaining multiple secondary features for precise identification and fuzzy recognition.We validated the validity and reliability of our model using a sensor dataset of a representative mobile phone and a sensor dataset attached to the human body.The advantage of our model is that the performance of the fused model architecture is superior tothe traditional single model architecture.