论文部分内容阅读
为了提高车载2D激光雷达对城市环境障碍物的分类能力与环境地图创建精度和无人车自主行为决策的安全性与准确性,提出了一种基于机器学习的环境特征分类方法。将2D激光雷达的观测数据帧分割为独立的数据段,每个数据段中包含一个环境障碍实体;在数据段的二维高斯概率密度空间中,以概率密度的等高线椭圆轴长、对数似然值和最大概率密度作为人工神经网络的样本数据元素,利用人工神经网络完成数据段分类;利用人工神经网络输出值的权重对分类的有效性进行判定,仅保留有效的环境特征,并对分类完成的观测数据进行特征提取。计算结果表明:在同一个试验场景中,当分类有效性判定条件被设定为分类稳定区间为[0.55,1],分类过渡区间为[0.45,0.55),分类无效区间为[0,0.45)的宽松条件时,共识别出98个环境特征,同一环境特征的多次观测数据的分类提取结果之间的最大标准差为30.7 mm,多个环境特征的平均标准差为5.1mm;当分类有效性判定条件设定为分类稳定区间为[0.65,1],分类过渡区间为[0.35,0.65),分类无效区间为[0,0.35)的严格条件时,共识别出93个环境特征,同一环境特征的多次观测数据的分类提取结果之间的最大标准差为22.0mm,多个环境特征的平均标准差为4.2mm,因此,提出的分类方法的噪声容忍能力强,分类精度高。
In order to improve the classification accuracy of urban environmental obstacle and the creation accuracy of environment map and the autonomous and autonomous decision-making of vehicle autonomous vehicles, a method of environmental feature classification based on machine learning is proposed. The data frame of 2D lidar is divided into independent data segments, each of which contains an environmental obstacle entity. In the two-dimensional Gaussian probability density space of the data segment, the elliptic axis length of the contour line of probability density Number likelihood value and maximum probability density are taken as the sample data elements of the artificial neural network, and the data segments are classified by the artificial neural network. The weights of the output values of the artificial neural network are used to judge the validity of the classification, and only the valid environmental features are retained The classification of the observed data for feature extraction. The results show that in the same experimental scenario, when the condition of classification validity is set as [0.55,1], the classification transition interval is [0.45,0.55], the classification invalid range is [0,0.45] 98 environmental characteristics were identified, the maximum standard deviation (SST) between the extracted results of multiple observations with the same environmental characteristics was 30.7 mm, and the average standard deviation of multiple environmental features was 5.1 mm. When the classification was valid When the condition of sexual judgment is set to the strict condition that the classification stability interval is [0.65,1], the classification transition interval is [0.35,0.65) and the classification invalid range is [0,0.35), 93 environmental characteristics are identified, and the same environment The maximum standard deviation (SD) between the extracted results of multiple observation data features is 22.0 mm, and the average standard deviation of multiple environmental features is 4.2 mm. Therefore, the proposed classification method has strong noise tolerance and high classification accuracy.