Abstract: In the natural scenario, it is difficult to extract a complete outline of the object because of background textures and light. This paper presents an object recognition method based on local contour feature. Firstly, the edge segments are split based on key points and merged based on Gestalt perception theory after the completion of extraction of the image edges. Secondly, it analyzes the contour feature made up of two or three adjacent segments. The semantic model of the contour feature that is made up of two adjacent segments is defined by the angle between the segments, the length and vectors of segments. The samples of local contour feature made up of two or three adjacent segments are drawn after the completion of decomposition of the template outline, and then the local contour feature made up of two or three adjacent segments is detected by combining the semantic model and matching the samples from the relations of the angle between the segments, the length and vector of segments. At last, the position points of detected local contour feature are obtained and grouped to recognize object. Experimental results show that the method can recognize the spacers and stock bridge dampers by processing the transmission line images acquired by UAV inspection.

Key words: object recognition, Gestalt perception theory, local contour feature, adjacent segments, group

摘要： 在自然场景下，受背景纹理及光线影响，很难提取到对象完整的轮廓，提出了一种基于局部轮廓特征的识别方法。提取图像的边缘线段，进行基于关键点的分裂与基于Gestalt感知定律的断续线段的合并处理，得到不同方位的独立直线段。分析2或3相邻线段构成的轮廓特征，通过线段之间夹角、长度和线段的向量对2相邻直线段构成轮廓特征的语义模型定义。分解对象的整体模板轮廓，手动绘出2或3相邻直线段所组成局部轮廓特征的样本，结合2相邻直线段语义模型与手动绘出的样本从线段夹角、长度以及对应线段向量之间的关系进行匹配，实现2或3相邻直线段组成局部轮廓特征的检测。求出所检测局部轮廓特征的位置点进行聚类实现对象的识别。通过对直无人机巡检所拍摄的输电线路图像处理，验证了这种方法能有效识别间隔棒和防震锤部件。

关键词: 对象识别, Gestalt感知定律, 局部轮廓特征, 相邻直线段, 聚类