基于智能手机的车辆行为实时判别与渐进矫正方法研究

发布时间：2018-01-12 04:36

本文关键词：基于智能手机的车辆行为实时判别与渐进矫正方法研究　出处：《计算机科学》2017年03期 　论文类型：期刊论文

【摘要】：目前基于智能手机的车辆行为识别算法存在着鲁棒性较差、识别率较低、无法应用于实时行驶判断等问题。针对上述问题,提出了基于智能手机的车辆行为实时判别与渐进矫正方法,以提高车辆行为识别的准确率和实时性。该方法利用车辆行为发生时存在的渐进变化数据来进行车辆行为的识别与渐进矫正分类,并通过采集过程数据作为分类器训练样本,提高支持向量机(SVM)分类器的车辆行为识别和预测能力。同时,针对传统滑动窗口检测的局限性,该方法采用了端点检测算法,从而能快速地从车辆行驶数据中截取并识别行为轨迹信息,以减少车辆行为的误判。实验结果表明,基于时间分段矫正的行为识别算法能够有效地对车辆行为进行预测,并最终达到较高的识别率,证明了该方法的有效性。
[Abstract]:At present, the vehicle behavior recognition algorithm based on smart phone has some problems, such as poor robustness and low recognition rate, which can not be applied to real-time driving judgment and so on. A method of real-time discrimination and progressive correction of vehicle behavior based on smart phone is proposed. In order to improve the accuracy and real-time of vehicle behavior recognition, the method uses the progressive change data when the vehicle behavior occurs to identify and classify the vehicle behavior. Through collecting process data as classifier training samples to improve the support vector machine (SVM) classifier vehicle behavior recognition and prediction ability. At the same time, aiming at the limitations of traditional sliding window detection. The endpoint detection algorithm is used in this method, which can quickly intercept and identify the behavior track information from the vehicle driving data, in order to reduce the misjudgment of the vehicle behavior. The experimental results show that. The behavior recognition algorithm based on time segment correction can effectively predict the vehicle behavior and achieve a higher recognition rate, which proves the effectiveness of the method.
【作者单位】：浙江工业大学计算机科学与技术学院;
【基金】：浙江省重大科技专项重大工业项目(2013C01112)资助
【分类号】：U495;TN929.53
【正文快照】： 到稿日期:2016-02-29返修日期:2016-05-30本文受浙江省重大科技专项重大工业项目(2013C01112)资助。1引言随着车联网研究的不断推进,车辆行为识别技术作为智能交通领域中场景理解的重要内容,成为车辆检测和跟踪之外的又一研究热点。车辆行为识别主要通过学习运动车辆的行为模

【相似文献】