基于改进JP算法的激光雷达可行驶区域检测

Drivable Area Detection Based on Improved JP Algorithm and Lidar

无人驾驶车辆对前方道路信息检测时,传统的基于欧式距离的密度聚类算法在计算密度不均匀的激光雷达数据时,在搜索精度和效率上存在一定的局限性。针对这一问题,提出一种改进的Jarvis-Patrick(JP)聚类算法。该算法通过k近邻(k-nearest neighbor)和共享最近邻SNN(Shared Nearest Neighbor)相似度间的关系来度量数据的局部密度选出代表点,对数据密度的变化具有伸缩性从而增加了算法的搜索速度和精度。对改进JP算法聚类后的簇进行评估,在道路边沿簇中使用随机抽样一致性算法(RANSAC)对两侧道路边沿点进行拟合。经实车实验表明,改进后的JP算法时间消耗上降低了32.6%,对被遮挡的道路边界及可行驶区域内障碍物检测精度均有提高。

When the unmanned vehicle detects the road information,the traditional Euclidean distance-based density clustering algorithm has certain limitations in the accuracy and efficiency of searching when dealing with the uneven density of the lidar data.To solve this problem,an improved Jarvis-Patrick(JP)clustering algorithm is proposed.The algorithm uses the relationship between the kNN(k-Nearest Neighbor)and SNN(Shared Nearest Neighbor)similarity to measure the local density of the data to select representative points,which is flexible for the change of data density,that increases the search speed and accuracy of the algorithm.The clusters after the improved JP algorithm clustering are evaluated,and the algorithm of random sample consensus(RANSAC)is used to fit the edge points of the roads on both sides of the road edge cluster.The actual vehicle experiment shows that the time consumption of the improved JP algorithm is reduced by 32.6%,the detection accuracy of obstacles in the occluded road boundary and the drivable area is improved.