Research On Location-awareness And Travel Mode Classification Based On Received Signal Strength Indicator

Roadside awareness is the core component of intelligent vehicle-infrastructure collaborative system,and the quality of awareness data will directly affect the application effect and feasibility in different scenario.Most of the existing roadside sensing scheme use conventional traffic sensors,such as intrusive detectors on the ground or suspended detectors that include support structures.These conventional detectors are expensive to install and maintain,and will seriously interfere with normal operation of urban traffic.The above deficiencies have led to the development of low-cost,easy-install and high-precision locationaware technologies for smart transportation data sensing.Among them,Wireless Sensor Networks(WSNs)has shown great promise for data collection,data analysis and data management.Among wireless location-aware technologies,the location method based on Received Signal Strength Indicator(RSSI)does not require modifications to the hardware and protocol layers of existing wireless sensing networks,and has the advantage of low cost and high accuracy,which has been widely used in the field of internet of vehicles.However,due to the loss of outdoor signal fluctuation,interference from traffic obstacles,deployment cost of roadside sensing units,and the insufficient accuracy of travel mode classification,there are still some urgent problems that need to solved for its large-scale integration applications in Cooperative Intelligent Transport Systems(C-ITS).This paper investigates the application of location sensing and travel mode classification algorithm in the intelligent vehicle-infrastructure collaborative system based on the collection and analysis of RSSI data,clarifies the differences between target localization in small-scale scenarios and large-scale scenarios,summarizes the existing problems and proposes countermeasures to improve the usability of location sensing and travel mode classification algorithm,and provides effective support for data collection,empirical analysis and management control of traffic information detection system in smart cities.The main research content and innovation points of this paper can be summarized as follows:(1)Fingerprint-based localization is a more suitable technical scheme for the demand of accurate localization of small-scale scenarios.However,traditional fingerprint localization method suffers from the problems of redundant data and longer training time.In this paper,we extend the local linear embedding method to semi-supervised learning machine,and propose a semi-supervised extreme learning machine localization method based on manifold dimensionality reduction by combining the vehicle driving characteristics in road environment.Experiments show that this method can provide ideal localization accuracy for vehicles with different driving speeds in sparse or dense sensor deployment environments with short training time and low sample size dependence.The method is suitable for small-scale non-line-of-sight scenarios such as semi-enclosed tunnels,underneath urban viaducts and dense buildings in urban canyons,and can support the realization of services such as blind zone warning,forward congestion alert and collaborative automatic cruise control.(2)Curve fitting-based localization is a more suitable technical scheme for the demand of fast localization of large-scale scenarios.However,traditional lognormal shadowing model suffers from the problems of high noise interference and poor fitting effect.In this paper,we propose a novel two-stage ranging-based localization method is proposed.Experiments show that this method can effectively suppress the fluctuating of wireless sensing signals in the road environment,reduce the peak of distance estimation error,and achieve better positioning performance while keeping low computational complexity.The method is suitable for smallscale non-line-of-sight scenarios with limited computing resources,more localization targets,more urgent time or small differences in propagation environments,and can support the realization of services such as crowd flow counting,vulnerable traffic participant warning and vehicle collision warning.(3)Neural network-based localization is a more suitable technical scheme for the demand of robust localization of large-scale scenarios.However,traditional neural network algorithm suffers from the problems of difficult model tuning and high computational complexity.In this paper,we improve the shortcomings of the current regression neural network optimization process and propose a ranging-based localization method by particle swarm optimization.Experiments show that this method can maintain good system robustness under different environments,speeds and equipment conditions.The method is suitable for large-range nonline-of-sight scenarios with sufficient computational resources,few localization targets,generous time or widely different propagation environments,and can support services such as crowd trajectory tracking,data-driven adaptive control of traffic lights and traffic health status.(4)In view of the low deployment cost and high real-time detection requirements of urban traffic information detection systems,passive detection based on WSNs is a more suitable technical solution.However,the existing travel mode classification algorithm suffers from the problems of high deployment cost and low screening accuracy.In this paper,we propose a travel mode classification system based on RSSI data.Experiments show that this system can effectively portray the real-time movement speed characteristics of users and accurately identify their travel mode.The method has the advantages of lower deployment cost,higher real-time performance and higher classification accuracy,and can support services such as cross-section and road condition monitoring and analysis,multi-traffic mode share monitoring and analysis,and regional O-D monitoring and analysis as a new generation of smart city traffic information detection system.In summary,for the localization and classification needs in different scenarios,this paper proposes fingerprint-based localization method for target accurate localization in small-scale scenarios,ranging-based localization method for target fast localization in large-scale scenarios,ranging-based localization method for target robust localization in large-scale scenarios and travel mode classification method based on single-point detection data.The four types of methods have their own characteristics and complement each other,covering noise filtering,distance estimation,location detection and travel mode classification,etc.They are assembled to form a set of algorithms with high accuracy and high real-time for the field of C-ITS,which provides a new scheme to realize the large-scale integrated application of vehicle-road cooperation technology in China.