Worker Productivity Monitoring System Based On Multi-type Inertial Sensor Cooperative Machine Learning

Limb activity recognition(LAR)technology based on inertial sensor is an emerging research direction in the field of human-computer interaction and pervasive computing and has been extensively studied in areas such as intelligent human-computer interaction,healthcare,education and motion analysis.However,the application of LAR technology based on inertial sensors in the construction field is still in a process of slow development and exploring the feasibility.Early studies failed to take into account the impact of inertial sensor location on construction activity recognition.And due to the complexity of construction activities,the recognition of construction activities based on inertial sensors remains to be studied.Therefore,the paper focuses on research the recognition of construction activity and workers' productivity under the cooperation of multiple types of inertial sensors.The main work is as follows:Construction activity recognition based on multi-type inertial sensor coordination.The literature research recognized four activities of the construction task decomposition of the loading and unloading materials as the object of construction activity recognition,and selected the four parts of the body to place the equipment unit through the correlation analysis between the activity and the body parts.Experiments 1 and 2 were performed to collect the loading and unloading construction activity data of the subject.The raw data was denoised by low-pass filter,signal segmentation and feature extraction to form feature vectors.In order to determine the window size,neural network classification methods are used to generate different location and category data models under five window sizes.The comprehensive analysis determines that the single location and category inertial sensor model has the best performance and the balanced signal segmentation window size is 3s.Subsequently,the experimental comparison and analysis of the single location,two location combinations,three location combinations and four locations inertial sensor category configuration model performance results,and finally obtains the optimal inertial sensor category and location combination as the ankle + upperarm + wrist acceleration(A).It proves the effectiveness of worker productivity recognition under the cooperation of multiple types of inertial sensors.Based on the optimal inertial sensor category and location combination,the actual construction activity data collected by experiment 2 was used to analyze and verify that the performance of the ankle-A +upper arm-A + wrist-A model is optimal.On the basis of the best model,the construction activity is recognized and the proportion of time for each type of construction activity is calculated.Compared with the actual results,the ratio of construction activity time recognized by the high-performance inertial sensor category and location combination model is not much different.The convincing data results validate the effectiveness of worker productivity recognition based on multi-type inertial sensors.Finally,a set of dynamic recognition and productivity monitoring system for construction workers' body postures under the cooperative work of multi-inertial sensors based on machine learning was developed and the reliability of the system was verified by actual data.