基于3D直方图与爬山法的K-means车灯零件检测算法

K-MEANS ALGORITHM OF AUTOMOBILE LAMP PARTS INSPECTION BASED ON 3D HISTOGRAM AND CLIMBING HILL

由于传统K-means聚类算法存在对初始值敏感和易陷入局部最优的缺陷,为了提高复杂性高的车灯零件分类的准确率,提出一种基于3D直方图与爬山法相结合的改进K-means的车灯零件检测算法。首先根据彩色车灯零件在颜色空间分布进行局部等间隔量化,得到自适应调节分辨率的3D颜色直方图。然后采用爬山法寻找局部等距量化的3D颜色直方图的极大值,根据极大值的特征动态地确定初始聚类数k和聚类中心。最后利用实际的车灯零件图像进行验证性实验。实验表明,该算法能准确检测出复杂性高的车灯零件,具有较好的稳定性和适用性。

Since traditional K-means clustering algorithm is sensitive to initial value and easy to fall into local optimum,in order to improve the accuracy of complex automobile lamp parts classification,a new K-means lamp parts inspection algorithm based on the combination of 3D histogram and climbing hill is introduced.First,the local equidistant quantisation is conducted according to the distribution of colour lamp parts in colour space,and 3D colour histogram with self-adaptive resolution is obtained.Then,the hill-climbing algorithm is employed to find the maximum value of 3D colour histogram which is locally quantified equidistantly,according to the characteristic of maximum value the initial clustering K and the clustering centre are adaptively determined.Finally,the practical lamp parts images are used to carry out validating experiment.Experiment demonstrates that the algorithm can accurately detect the complex lamp parts with fairly good stability and applicability.