Research On The Recognition System Of Human Lower Limb Behavior Based On Edge Computing

In the context of the era when the cloud computing model that centrally processes data has not yet adequately responded to the explosive growth of massive application data,an edge computing model that can reduce the computing burden of the cloud center platform and endow edge devices with intelligence has emerged.With the widespread application of human behavior recognition technology,a large number of algorithm models are required.Traditional training methods often consume a lot of computing resources and labor costs.Combining the framework of edge computing,this paper studies a system for the recognition of human lower limb behavior characteristics,proposes a cloud-side-end collaborative algorithm training method,avoids large-scale centralized training methods,and focuses on algorithm personalization and migration learning.The cloud conducts universal model training,the edge gateway conducts migration learning to achieve personalized training,and the edge uses personalized models for identification,providing different testers with personalized models.This paper uses EdgeX to design the edge gateway in the edge computing framework.At the same time,it designs an edge device that collects two kinds of limb behavior signals in real time and can recognize multiple human lower limb movement patterns.The main research contents include:(1)In the era when the cloud computing model of centralized data processing has not been able to respond well to the explosive growth of massive application data,an edge computing model that can reduce the computing burden of the cloud center platform and endue the edge devices with intelligence arises at the right moment.Combined with the new paradigm of edge computing concept,this paper builds edge gateways through the open source edge computing framework Edgex,configures different micro-service functions in each service layer inside Edgex,completes the service design on the south and north sides of the edge gateway,and coordinates the information interaction of each micro-service within the Edgex framework.and realize the multi-device management control and data exchange for the devices at the edge end.(2)This paper designs a peripheral human lower limb behavior recognition device,which integrates the human surface EMG signal acquisition circuit,signal conditioning circuit,foot signal detection circuit,inertial sensor signal acquisition interface and the peripheral main control platform.The digital filtering of sEMG and IMU signals,the data segmentation based on sliding time window and the extraction of different data features from the two signals are realized respectively,and the signal preprocessing is completed.(3)At the level of algorithm design,this paper uses a branched convolutional neural network to fuse two kinds of lower limb movement signals to realize the recognition of four different lower limb movement patterns:walking,going down stairs,jumping,and going up stairs,and obtains a CNN with a certain generalization ability.A universal model of human lower limb behavior recognition.(4)In order to realize the personalized recognition of different recognition objects,this paper adopts a model-based transfer learning method,performs transfer learning on the basis of a universal model and an open source gesture recognition model,and adopts the transfer learning operation of freezing part of the network layer and fine-tuning,and the training is obtained.Personalized model.Compare and analyze the model effect of transfer learning.(5)Design a verification experiment for model recognition at the edge,and at the same time access the gateway to recognize the behavior of the lower limbs,and display the recognition results in real time at the edge or the edge gateway respectively,verify that multiple devices are connected to the gateway and use a personalized model for identification,complete System construction.