基于单目视觉测量技术的车辆尺寸识别方法

A Method for Recognizing Vehicle Dimensions Based on Monocular Vision Measurement Technology

为了提高车辆尺寸识别效率与稳定性,同时有效降低图像信息处理量与硬件成本,进而实现车辆等效荷载的快速获取,基于计算机图像像素尺度因子提出了一种单目视觉车辆尺寸测量方法以实现目标图像中车辆尺寸的快速获取。基于摄影相机成像原理推导了计算机图像纵向像素尺度因子表达式,结合目标图像序列中的结构化信息对模型参数进行了优化。利用图像轮廓线与RGB色度空间像素定位算法实现了目标车辆纵向端底部像素坐标的提取。对3种不同尺寸模型车与真实交通流依次开展了尺寸识别试验,以验证该方法测量精度与实用性。结果表明:利用本方法得到的模型车尺寸误差在14.5%以内,平均误差仅5.3%;针对真实交通流中14辆任意汽车的尺寸识别,本方法最大误差为8.4%,平均误差4.8%,证明了本研究算法对不同的目标车辆识别效果较好;与现有其他基于机器视觉的车型识别技术相比,该方法综合设备参数化信息与场景空间信息,在目标车辆的尺寸识别具有明显的优势,由于不需要进行复杂的图像处理步骤,几乎不占用场地,可提升实际交通流监测的识别稳定性与实用性。该方法也可为车辆尺寸识别提供一种新的求解思路,进而为智能交通系统中车型识别提供准确与可靠的基础数据。

In order to improve the efficiency and stability of recognizing vehicle dimensions,effectively reduce the image information processing amount and hardware cost,thus realizing fast acquisition of vehicle equivalent load,and realize rapid acquisition of vehicle equivalent load,a method for monocular vision measurement of vehicle dimensions based on pixel scale factor of computer image is proposed to achieve fast acquisition of vehicle dimensions in the target image.The expression of longitudinal pixel scale factor of computer image is derived based on the camera imaging principle,the model parameters are optimized combining with the structural information in the target image sequence.The pixel coordinates at the bottom of the longitudinal end of the target vehicle are extracted by using the image contour and the pixel location algorithm based on RGB chromaticity space.Three model vehicles with different dimensions are tested in real traffic flow to verify the measurement accuracy and practicability of this method.The result shows that(1)The dimension error of the model vehicle obtained by the proposed method is within 14.5%,and the average error is only 5.3%.(2)For the dimension recognition of 14 arbitrary vehicles in the real traffic flow,the maximum error of this algorithm is 8.4%,and the average error is 4.8%,which proves that the proposed method has good recognition effect on different target vehicles.(3)Compared with other existing vehicle type recognition technologies based on machine vision,the proposed method integrates equipment parameterization information and scene space information,and has obvious advantages in dimension recognition of target vehicles.As it does not require complex image processing steps and almost does not occupy space,it can improve the recognition stability and practicability of actual traffic flow monitoring.(4)This method can also provide a new solution for vehicle dimension recognition,and provide accurate and reliable basic data for vehicle type recognition in ITS.