路侧多传感器融合中的联合标定优化设计

Optimization Design of Joint Calibration for Roadside Multi-sensor Fusion

摘要:

在路侧感知中，交通场景复杂多变，单类传感器无法满足全天候、全天时工作的要求，多传感器融合逐渐成为发展趋势，而多传感器融合的前提是实现传感器之间的联合标定。在激光雷达和相机的共视区内摆放标定物，从而获得特征点的点云坐标和对应像素坐标，再使用EPnP（Efficient Perspective-n-Points）算法求得激光雷达和相机之间的外参；对激光雷达路面的点云进行上采样，根据联合标定的外参矩阵生成对应的像素坐标，然后采用决策树回归、随机森林回归算法训练模型；根据激光雷达标定的结果实现像素坐标到WGS84坐标的转换，该方法不需要将激光雷达和相机集成在一起。最后，经实验验证该联合标定方案的有效性。

Abstract:

In the roadside perception， the traffic scene is complex and changeable， and a single type of sensors cannot meet the requirement of all-weather and all-day work. Therefore， the multi-sensor fusion has gradually become a development trend， which is on the premise that the joint calibration among sensors should be achieved. In this paper， by placing markers in the common view area of lidar and camera， the point cloud coordinate and corresponding pixel coordinate of feature points are obtained， and external parameters between lidar and camera are obtained by EPnP （Efficient Perspective-n-Points） algorithm. The point cloud of the lidar on the road surface is up-sampled， and corresponding pixel coordinates are generated according to the matrix of the joint calibration. The model is trained to fit the point cloud and pixel by using the algorithm of decision tree regression and random forests regression， and then the pixel coordinate is converted to WGS84 coordinate according to the results of lidar calibration. This method does not need the integration of lidar and camera. The effectiveness of the proposed joint calibration scheme is verified by the experiment.