Behavior Recognition For Elevator Users Based On Machine Learning

At present,most of the researches on elevator(straight elevator)user behavior recognition are based on monitoring data in the elevator.With the rapid development of sensor technology,it is of great significance to use sensor data in portable external devices to conduct behavior recognition research on elevator users.The combination of various sensor data and machine learning algorithms for behavior recognition has a huge advantage in behavior detection.The monitoring goal of elevator user behavior recognition is to be able to detect the behavior of people in the elevator in order to promptly remind elevator users to stop dangerous actions.This is a key step to realize intelligent monitoring and use for elevators.The paper mainly studies the use of elevator user behavior data recorded by the sensor module,such as acceleration sensor data,gyroscope data and other motion data to identify the elevator user behavior in the elevator scene.First of all,this article develops an application to acquire sensor data and complete the behavior data collection of elevator users in the elevator scene.After analyzing and processing the collected sensor data,a characteristic data set is constructed.Finally,the elevator user behavior recognition algorithm is used to build a classification model for behavior recognition and make corresponding voice reminders.The details are as follows:This article develops an application that collects sensor motion data.Fix the mobile device with this application on the elevator floor for data collection.Elevator user behaviors include: normal behaviors,such as standing,squatting,and sitting normally;stomping behaviors,such as stomping on different positions on the elevator floor;and behaviors of hitting elevator walls,such as hitting on elevators in different directions.After completing the data collection,analyze and preprocess the collected sensor data.Feature extraction is carried out according to the characteristics of the data,and the feature data set is constructed.In this scenario,according to the features of the data and the characteristics of the classification model,a parameter optimization scheme is designed.Find the appropriate number of random forest decision trees and the number of split features,and then optimize the parameters.After the training model is obtained,the test set data is identified,and an elevator user behavior recognition algorithm is constructed.Then use weighted voting to participate in the decision,and select the category with the largest weighted value as the final output of the recognition algorithm.Use the elevator user behavior recognition algorithm to experiment,the experiment results show that the random forest optimization algorithm proposed in this paper has the best classification performance: the accuracy is 94.97%,accuracy rate is 94.48%,recall rate is 92.03%,F1-score is 0.92,the generalization error is 0.052793,the algorithm achieves very accurate identification.Then,a comparative experiment is conducted on the role of single sensor and multi-sensor in the behavior recognition.The experiment results show that the behavior recognition effect based on the multi-sensor is better.Finally,a voice reminder module is developed,which transmits the predicted value of the recognition algorithm to the mobile terminal.Then call the voice reminder module to output the corresponding voice reminder of the dangerous action.