基于智能手机的车辆行为实时判别与渐进矫正方法研究

Real-time Determination and Progressive Correction of Vehicle Behavior on Smartphones

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。

Up to now, the relevant research has some drawbacks：poor robustness, low accuracy rate and non-real time. To solve these problems,a vehicle behavior recognition algorithm of real-time determination and progressive correction on smartphone was proposed. This algorithm classifies vehicle behavior by the data generated during driving process, and uses the collected data as training samples to improve recognition and prediction capability of SVM. For the limita- tions of traditional sliding window, the endpoint detection algorithm is used to quickly extract useful information from the complete vehicle behavior,which reduces misjudgment simultaneously. The experimental results show that correc- tive algorithm on time-based segmentation can effectively predict the vehicle behavior, and ultimately achieve high re- cognition rate, which demonstrates the effectiveness of this method.