Research of Vehicle Detection and Classification in Wireless Sensor Network
|School||Dalian University of Technology|
|Course||Communication and Information System|
|Keywords||Wireless sensor networks Vehicle Detection Target recognition Decision Fusion|
In military reconnaissance and ground reconnaissance concealment and camouflage complex topography uninterrupted automatic reconnaissance and surveillance functions and expand the scope of information to detect spatial and temporal, the developed countries have attached great importance to its research and application. In recent years, the development of wireless sensor network has a small size, its random nature, is very suitable for the the ground enemy situation reconnaissance troops, equipment and supplies monitoring purposes, monitor vehicles and their detection and recognition is one of the important applications. Strong sound and vibration signals generated in the process of the vehicle, you can take advantage of the sound and vibration sensors to monitor the vehicle, so the domestic and international wireless sensor network monitoring system usually contains two sensors. The need for monitoring military vehicles, wireless sensor networks are usually deployed in the natural environment is more severe in remote areas affected by the natural conditions, the vehicle signal detected by the sensor nodes will contain a lot of interference noise, how these noisy signal extract the useful signal of the vehicle and its make accurate detection and identification is a very challenging task. The context of wireless sensor network monitoring vehicle, the use of sound and vibration signals generated when the vehicle is in motion, the vehicle target detection and identification of the monitoring area contains the following sections: detect whether the target signal, the target signal preprocessing , extracting a signal feature, the extracted feature into the classifier for classification, and a plurality of nodes, the decisions made by the same target at the same time to do a global fusion. To meet the special needs of the sonic booms signal detection in wireless sensor networks in vehicles, combined with the energy detection method of the time domain and frequency domain select the maximum power detection method proposed CFAR detection based on a constant false alarm rate and spectral distribution of the double threshold detection algorithm, which can be subject to serious noise pollution signal extracted vehicle sonic booms signal. Then the vehicle signal detection using wavelet packet denoising and downsampling, complete the pre-processing of the signal. After pretreatment of the signal, respectively, using FFT-based feature extraction method and the target signal characteristics based on wavelet packet analysis feature extraction methods, and the extracted features into the neighbor classifier and support vector machine classification recognition results to obtain a single node. Wireless sensor network monitoring objectives, the deployment of a large number of sensor nodes in the monitoring region, there will be multiple nodes simultaneously to detect the target and make the results of the identification when the target appears. This feature using wireless sensor networks, this paper proposes a fusion algorithm based on multi-node global energy decision-making and global integration of multiple nodes to make decisions, to get the final result of the network of the target. In order to assess the performance of the algorithm in this article, we use data from the DARPA SensIT project team to do the real wireless sensor network experiments, which contain sound and vibration signals generated when a large number of tracked vehicles and wheeled vehicles. By comparing the experimental results with the relevant references, indicating that the vehicle identification of the algorithm for a wireless sensor network is valid.