基于DSP的SVM算法实现鱼眼镜头畸变校正

鱼眼镜头应用于全方位视觉跟踪导航系统中,但其产生的畸变如不加以精确实时的校正必然导致定位错误和偏差。本文创新地采用鱼眼镜头所采集的图像特征点训练支持向量机(SVM),使其产生回归径向距离的模型并生成坐标映射,然后通过查表法实时的校正采集的图像,用TMS320DM642处理鱼眼镜头畸变的实时校正。结果证实,该方法校正效果好,响应速度快,满足图像处理的实时性要求;并且用DSP代替以往PC运算,可简化整体结构,压缩系统体积。

Real-time rectification of fisheye lens using SVM based on DSP

Fisheye lens is used in omni-directional image system of tracking and navigation.The distortion of fisheye lens,however,must lead to error in positioning if isn't accurately corrected.In this passage,SVM is trained using points gathered by fisheye lens and then makes a model for radial distance regression.Then the distortion can be real-timely corrected by means of table checking.The whole rectification process is finished on DSP TMS320DM642.Results using this method have verified that the degree of the rectification is sufficient and provides high speed response,which is required for accurate analysis of the image;instead of PC,DSP can simplify the structure and reduce the size.