Abstract: According to the characteristics of low speed and high slip ratio of construction vehicles，a method was proposed based on binocular sequence images in order to quickly detect the relative positions and actual speeds of the vehicles. The vehicle-borne binocular vision sensors were used to collect the image sequences of the surrounding environments continuously. And the distances from the environment feature points to the origins of the camera coordinate systems were calculated through matching the captured each frame binocular images using speeded up robust features(SURF)，so as to achieve vehicle relative locations. Meanwhile, the feature points of two adjacent frames were tracked and matched based on SURF，and the matching feature points were divided into far points and near points according to different depths of the fields . The rotation matrix and translation vector of the coordinate systems in vehicle motions were calculated by using the far points and the near points respectively, and then optimized them by Levenberg-Marquardt method. Finally, the velocities of the vehicles were estimated according to optimized rotational matrices and translation vectors. The results of outdoor simulation test show the effectiveness and feasibility of the proposed method.

Key words: binocular vision, image sequence, feature tracking and matching, ego-velocity estimation

摘要： 针对工程车辆行驶速度低、滑转率高的特点，提出了一种基于双目序列图像的检测方法，以便快速检测工程车辆的相对位置与实际行驶速度。将双目摄像机安装在车辆上，连续采集周围环境的序列图像；利用SURF(speeded up robust features)特征对已采集到的各帧双目图像进行立体匹配，计算出环境特征点到摄像机坐标系原点的距离，从而实现车辆的相对定位；再对相邻两帧图像进行特征跟踪匹配，根据不同景深将匹配特征点对划分为远距点对和近距点对，分别利用远距点对和近距点对估算车辆运动过程中坐标系的旋转矩阵和平移矢量，并利用Levenberg-Marquardt法进行优化求解；最后根据优化后的旋转矩阵和平移矢量计算出车辆的行驶速度。户外模拟试验结果表明了方法的有效性和可行性。

关键词: 双目视觉, 序列图像, 特征跟踪匹配, 自车速度估计