The Research And Implement Of Moving Vehicle Recognition Based On Acoustic Sinal

This paper focuses on motor vehicle recognition based on acoustic signal, using thecharacteristic of acoustic signal to realize classification of heavy and light vehicle. Thisthesis mainly includes: acoustic target detection, feature extraction, classificationalgorithm design and implementation of DSP. Following are the main content of thisarticle:(1) A study on sound algorithm of target detection. Due to the environmental noiseand attenuation, it is difficult to detecting target. In order to solve this problem, thisarticle adopts the rate Cell Average Constant False Alarm Rate (CA-CFAR) detectiontechnology, which can reduce the loss of signal-to-noise ratio and improve the detectionperformance. Combined with the actual algorithm parameters, simulation are carried on.The result shows that when the condition meets the requirement of signal-to-noise ratio>4db, the false-alarm probability pf25%, the detection probability pd can reach90%,which can meet the needs of the project.(2) Acoustic feature extraction algorithms are discussed. This article adopts featureextraction algorithm based on the Fast Fourier Transfer(FFT)、the Hilbert-Huangtransfer(HHT) and the wavelet packet transform. Wavelet packet feature extractionalgorithm was optimizited by using two kinds of feature selection methods. And aimproved wavelet packet feature extraction algorithm was put forward: the bottomwavelet packet energy feature extraction algorithm and the optimized wavelet packetenergy feature extraction algorithm. The simulation results show that the improvedalgorithms have smaller dispersion and less feature dimension. So it is suitable to beused in this project.(3) In terms of classifier design, study on the common classifier: KNN classifier,bayes classifier, support vector machine (SVM) classifier and fuzzy classifier. Thisarticle uses the hierarchical ideas to improve traditional fuzzy classifier, which can solvethe problem of too much classification rules and low efficient. The simulation resultsshow that the hierarchical fuzzy classifier algorithm not only has higner classificationrecognition rate, but also has stable performance and good robustness. (4) This paper adopts the OMAP-L138dual-core processors DSP kernel runningon the DSP/BIOS real-time operation system to apply the selected algorithms anddesigns the communication solution between ARM and DSP. The experiments verifythat the performance of recognition system is stable, real-time processing and achievingrecognition rate in line with the indicators.