Abstract: The contour corner detection and feature construction are the key technology in Rotation and Scale Invariant（RSI） target recognition based on contour corner. Aiming at the insufficiency of the existing corner detection in precision and anti-noise capacity, this paper presents an improved contour corner detection method, and constructs a characteristic string based on target’s principal axis and contour corners, using the dynamic programming algorithm to identify the target. In this algorithm, the rotation angle of the principal axis of target is regarded as the angle of target’s attitude to calculate. Experiments prove that this algorithm can rapidly identify the rotation angle of the target, and classify the target. The algorithm also has translation invariance, rotation and scale invariance and well anti-noise capacity.

Key words: contour corner, principal axis, rotation and scale invariant, multi-target recognition, rotation angle

摘要： 轮廓角点检测与特征构造是基于轮廓角点的RSI多目标识别算法的关键。针对现有的轮廓角点检测方法在准确性与抗噪能力的不足，提出一种改进的轮廓角点检测算法，构造一种基于目标主轴与轮廓角点的特征串，利用动态规划算法计算特征串间的相似度进行目标识别。算法中把目标主轴的旋转角度作为目标的姿态角。实验证明该算法能够快速地识别出目标的旋转角度，对目标进行分类，且具有平移不变性、旋转不变性、尺度不变性以及较好的抗噪能力。

关键词: 轮廓角点, 主轴, 旋转与尺度不变性, 多目标识别, 旋转角度