| Intelligent Traffic Systems(ITS)aim to provide convenient travel services for citizens and lay the foundation for sustainable urban development.The effectiveness and accuracy of traffic data collection(TDC)is a key means to improve the intelligence level of ITS.Therefore,the research on TDC technology is of great significance.Traditional TDC equipment includes cameras,radars,ultrasonic waves,etc.,while the performance of these TDC equipment are easily affected by environmental factors such as light,temperature,humidity,etc.,resulting in a decline performance at night,rainy,foggy and other adverse weather conditions.To address these issues,the wireless sensing-based traffic data collection(WTDC)method has become a research hotspot in recent years due to its advantages of low deployment cost,non-invasiveness and ubiquity.However,the existing WTDC methods have limited applicability and low recognition performance,which hinders the collection of traffic data.In addition,existing WTDC methods focus on leveraging the energy attributes of wireless signals with a frequency of 2.4 GHz,while ignoring the application of Sub-GHz bands.In addition,the antenna height and carrier frequency determine the energy attributes of wireless signals.So far,few works explore the relationship between the antenna height,carrier frequency and WTDC performance.Therefore,it is necessary to clarify the relationship among antenna height,carrier frequency and WTDC performance.To clarify the above relationship,driven by the powerful feature learning ability of machine learning technology,this paper proposes a WTDC method based on machine learning to clarify the relationship among antenna height,carrier frequency and WTDC performance.The main works of this paper are as follows:1)This paper develops a five-category dataset based on received signal strength(RSS)data.This dataset contains wireless energy data for three carrier frequencies(433 MHz,915 MHz and 2.4 GHz)and five antenna heights(0.4 m,0.8 m,1.2 m,1.6 m and 2.0 m).The dataset consists of 2000 samples:one-pedestrian,multi-pedestrians,one-bicycle,multi-bicycles and one-car,respectively indicating that there is one pedestrian,multiple pedestrians,a bicycle,multiple bicycles and a car in the wireless sensing area.The dataset is publicly available at https://github.com/TZ-mx/Wi Param.2)Aiming at the problems of limited applicability and low recognition performance of existing WTDC methods,this paper designs a convolutional neural network(CNN)model based on attention mechanism,named wireless sensing-based lightweight attention machine learning(Ws-LAML).Ws-LAML takes RSS data as input,since RSS contains the useful temporal feature of the gait for TDC.Ws-LAML consists of several lightweight CNN modules in series and a convolutional block attention mechanism(CBAM)to learn high-level features from RSS signals.Experimental results on the developed five-category RSS dataset show that the proposed Ws-LAML not only achieves higher accuracy,but also exhibits lower computational complexity compared with state-of-the-art TDC methods.3)To clarify the relationship among antenna height,carrier frequency and TDC performance,and to clarify the multi-parameter influence mechanism of the wireless sensing method for TDC,this paper uses the Ws-LAML model to conduct experiments on the developed dataset.The results show that both Sub-GHz and 2.4GHz wireless energy data can lay the foundation for the construction of wireless sensing methods for TDC,and Ws-LAML obtains the best performance with an average accuracy of 98.8% at 2.4 GHz band and an antenna height of 0.8 m on TDC tasks.In summary,this paper aims to propose a low-cost and high-precision TDC method,so as to clarify the relationship among antenna height,carrier frequency and TDC performance.In addition,this paper also develops a five-category dataset based on RSS data.In the future research on TDC methods,the multi-task learning method of multi-sensor fusion can be used for more in-depth research. |
