基于ResNet和Chan算法的行车环境声音感知  

Sound Perception of Driving Environment Based on ResNet and Chan Algorithm

作　　者：朱振飞 葛动元 姚锡凡[2] ZHU Zhen-fei;GE Dong-yuan;YAO Xi-fan(School of Mechanical and Automotive Engineering,Guangxi University of Science and Technology,Liuzhou Guangxi 545006,China;School of Mechanical and Automobile Engineering,South China University of Technology,Guangzhou Guangdong 510640,China)

机构地区：[1]广西科技大学机械与汽车工程学院,广西柳州545006 [2]华南理工大学机械与汽车工程学院,广东广州510640

出　　处：《计算机仿真》2024年第9期121-126,161,共7页Computer Simulation

基　　金：国家自然科学基金(51765007)。

摘　　要：随着自动驾驶技术的飞快发展和道路交通安全的要求,车辆对环境信息感知的要求也越来越高。道路声音的感知是自动驾驶走向成熟的重要组成部分,具有广泛应用前景。利用ResNet18网络模型对特定声学事件进行识别,给数据集添加不同方差的噪音以模拟真实行车环境,并联合多层感知机模型和Chan算法进行定位,提出了一种Net-Chan定位方法,将时延估计值输入网络模型。通过Chan算法解算得到最终目标位置,实现了自动驾驶车辆行车环境的声音感知,具有较高的识别准确度,提高了定位精度和抗噪性,可作为视觉系统在受环境限制时的环境信息的收集补充与融合。With the rapid development of autonomous driving technology and the increasing demand for road traffic safety,vehicles have higher and higher requirements for the perception of environmental information.The perception of road sound is an important part of the maturity of autonomous driving and has a wide application prospect.In this paper,ResNet18 network model is used to identify specific acoustic events,noise with different variances is added to the data set to simulate the real driving environment,and multi-layer perceptron model and Chan algorithm are combined for positioning,a Net-Chan positioning method is proposed,and the time delay estimate is input into the network model.By using the Chan algorithm to calculate the final target position,the sound perception of the driving environment of autonomous vehicles has been achieved,with high recognition accuracy,improved positioning accuracy and noise resistance.It can be used as a supplement and fusion of environmental information for visual systems when limited by the environment.

关 键 词：自动驾驶 残差网络 声音识别 声音定位 

分 类 号：TP273.5[自动化与计算机技术—检测技术与自动化装置]