基于自适应栅格聚类的轻量化车辆检测算法

Lightweight Vehicle Detection Algorithm Based on Adaptive Grid Clustering

提出了一种自适应栅格聚类的轻量化车辆检测算法。采用半径滤波和最小二乘法对原始点云进行噪声点去除和地面拟合处理,采用最大最小值栅格化和自适应栅格聚类算法生成若干个聚类目标,使用多层感知机网络对聚类目标进行二分类。在KITTI数据集上进行训练和验证算法实验,结果表明:算法在不同场景下具有环境适应性,与其他3D检测算法相比,车辆识别准确率平均提高了7.95%。

A lightweight vehicle detection algorithm based on an adaptive grid clustering was proposed.Radius filtering and the least square method were used to remove noise points from the original point cloud and perform ground fitting processing.Maximum and minimum grid transformation and an adaptive grid clustering algorithm were employed to generate several clustering targets.A multi-layer perceptron(MLP)network was utilized to perform binary classification of the clustering targets.Training and algorithm validation experiments were conducted on the KITTI datasets.The experimental results show that the algorithm has environmental adaptability in different scenes.Compared with the other 3D detection algorithms,the proposed algorithm improves the average accuracy of vehicle recognition by 7.95%.