An on-board vision system is recognized as a promising tool for vehicle early warning and monitoring. Timely accurate estimation of vehicle speed is critical in allowing the on-board vision system to calculate the vehicle location, plan a driving path, and apply emergency brakes to avoid accidents. However, the scene images captured by the vision system always suffer from global motion blur, which causes great difficulty in precisely estimating vehicle speed. While extensive efforts have been focused on blurred image restoration and real-time driving speed estimation in highway scenarios, very limited work has addressed urban scenarios in which the vehicle speed is often less than 40 km h⁻¹. In order to bridge this research gap, this study proposes a new method for real-time vehicle speed estimation. Firstly, the spectrum characteristics of blurred images at low vehicle speeds were investigated to determine the relationship between the direction and spacing of the spectrogram and vehicle motion parameters. Then, the blur-direction and blur-scale of the vehicle motion were analyzed by double Radon transform to develop a speed estimation model. Experimental evaluation results demonstrate that the proposed method was able to estimate vehicle speed in urban scenarios without updating the hardware of existing on-board vision systems. The estimation error was below 7.13% and the calculation efficiency of a single frame was 30 ms, both of which meet the practical application requirements of intelligent vehicles.

车载视觉系统被认为是用于车辆预警和监测的有前途的工具。及时准确地估计车速对于使车载视觉系统能够计算车辆位置，规划行车路线并应用紧急制动以避免事故至关重要。然而，由视觉系统捕获的场景图像总是遭受全局运动模糊，这在精确估计车速方面造成很大困难。尽管在高速公路情景中已集中精力进行模糊图像恢复和实时行车速度估计，但针对城市情景的工作却非常有限，在这些情景中，车速通常低于40 km h ^-1。为了弥合这一研究差距，本研究提出了一种实时车辆速度估计的新方法。首先，研究了低车速下模糊图像的光谱特征，以确定光谱图的方向和间距与车辆运动参数之间的关系。然后，通过双Radon变换对车辆运动的模糊方向和模糊度进行分析，以建立速度估计模型。实验评估结果表明，该方法能够在不更新现有车载视觉系统硬件的情况下估计城市场景中的车速。估计误差在7.13％以下，单帧计算效率为30 ms，均满足智能汽车的实际应用要求。