应用弦切变换提取几何特征实现目标检测

提出了应用弦切变换(CTT)来提取目标几何形状特征的方法,并基于对提取形状特征的匹配实现了对二维目标的检测。介绍了弦切变换的理论依据和算法的实现过程,同时进行了误差及可靠性分析。分析和实验表明,由于CTT可基于边缘点信息提取出更有意义的几何形状特征,这种几何特征不但具有平移、旋转和缩放的不变性,同时还可以输出目标相对于模板的旋转角度、缩放尺度等运动参数信息,因此,该方法在一定程度上克服了传统方法中灰度特征易受复杂环境和光照变化等影响的缺点,同时该特征对于目标边缘的部分失真或缺损也具有一定的鲁棒性。最后,对多组复杂环境下的图像序列进行了仿真实验,实验结果显示,检测的平均成功率达90%以上,而且在目标缺失或变形达到40%时,仍能得到比较准确的检测结果,这些结论验证了该方法的有效性。

Object detection based on geometric feature extracted by chord-tangent transformation

A method to extract the geometric features of planar objects was proposed by using the Chord-tangent Transformation (CTT), then a 2D object was detected based on matching the geometric features obtained by the CTT. The theoretical basis of the CTT was introduced and how to realize the method was described. Meanwhile, the errors and reliability of the method were analyzed and discussed. It was pointed out that the CTT can extract a kind of senior geometric feature from the information of edge points, and the extracted geometric feature not only has invariant characters for translation, rotation and scale, but also can obtain some important parameters such as rotation angles and movement levels. Therefore, this method solves many problems caused by the unsteadiness of gray features in complex environments and changing illumination. Moreover, it shows a robustness for the condition that there is distortion or damage in the object edges. Finally, an experiment was undertaken with some image sequences from various complex conditions, and the results show that the average accuracy of object detection is over 90%. Even if the distortion or damage of the object edges has been more than 40%, it also can offer a more accurate detection result. These results prove the effectiveness and accuracy of method.