Research And Application Of Sensor Node Of Transportation IoT Based On Data Fusion

As we know, sensor network has some problems of weak sensor ability, single sensor way, uncomprehensive sensor information, etc. In order to enhance sensor node’s sensor ability and remove information isolated island, multiple sensors are detected in parallel, all data will be processed by the method of data fusion. On the one hand, multiple sensors working cooperatively can make up each shortages in sensor information and improve environmental adaptability. On the other hand, as the data detected by each sensor has its own characteristics, fusing different kinds of data can also improve the detection accuracy, efficiency and coverage.This paper mainly apply multi-sensor data fusion in vehicle recognition, and data is from loop, geomagnetic and video detectors. To ensure space-time synchronization,a method combining software with hardware is used in this paper. In the hardware,arrange all the sensors reasonably to guarantee space synchronization of multi-source data; in the software, get the data based on time to maintain synchronized time.According to the characteristics of all kinds of data, pre-process these data and extract corresponding features. A novel method called IPSO-MSK-RVM is adopted to fuse the three kinds of features to recognize vehicle type. The method apply RVM to achieve multi-class classification and utilize PSO to optimize the parameters of multi-kernel used in RVM.Firstly, as RVM is essentially a binary classifier, in order to recognize multiple vehicle types, this paper use one-against-one method which is applied most widely. In fact, one-against-one needs many binary classifiers leading huge computation. In view of the defect, a new one-against-one based on RVM is presented which significantly improve classification speed and ensure classification accuracy at the same time. Secondly, the key of RVM is kernel function and our data is heterogeneous, so only use single kernel function cannot classify reasonably. Instead,multi-kernel model is suitable for heterogeneous and complex data, while without the support of theory, its parameters are often set by experience or experiments.In this case, a modified PSO algorithm is presented to get global optimum parameters of multi-kernel, which improve the accuracy of vehicle recognition effectively.Finally, to put IPSO-MSK-RVM into practice, a sensor node of multi-sensor parallel detection and real-time fusion is designed and the corresponding vehicle recognition system is developed. Experiments show that IPSO-MSK-RVM is effective on vehicle recognition and has good practicability.