基于SSD和改进双目测距模型的车辆测距方法研究

Research on vehicle ranging method based on SSD algorithm and improved binocular ranging model

针对汽车辅助驾驶系统中使用双目相机对前方车辆测距精度较低的问题,提出一种基于SSD车辆检测和改进双目测距模型的车辆测距方法。该方法分为3个阶段。第一阶段,使用深度学习多尺度特征的SSD算法快速准确地检测定位车辆;第二阶段,使用SURF+RANSAC算法匹配校正后的左右车辆的特征点;第三阶段,在分析双目相机测距误差的基础上,构建新的测距模型,并采用遗传算法对测距模型中的参数进行非线性拟合,估计出参数的最佳值,最后使用改进的双目测距模型实现更准确的测距。实验结果表明,在40 m的距离范围内,车辆正确识别率为94.32%,每幅图像的车辆检测平均耗时为50.18 ms,所提方法的测距误差控制在2m以内,平均误差率为2.36%,较改进之前的方法精度提高了6.19%,具有较好的实用价值。

In order to fix the problem of low binocular ranging accuracy of front vehicle in assisted driving,a vehicle ranging method based on SSD algorithm and an improved parallel ranging model is proposed.The method is divided into three stages.In the first stage,a deep learning SSD algorithm is used to detect and locate vehicle quickly and accurately.In the second stage,the corrected feature points of left and right vehicles are matched by using SURF and RANSAC algorithm.In the third stage,a new binocular camera ranging error is analyzed to construct a new ranging model,and the genetic algorithm is used to fit the parameters in ranging model non-linearly,and the best value of the parameters are estimated.More accurate ranging is achieved with improved binocular ranging model.The experimental results show that the correct vehicle recognition rate is 94.32%,the average vehicle detection time of per image is50.18 ms and the distance measurement error of proposed method is within 2 m for 40 m ranging,and the average error rate is 2.36%,which means 6.19%higher accuracy than previous method is achieved and proposed method has good practical value.