基于DSP的直线特征快速提取算法研究

Research on rapid linear feature extraction algorithm based on DSP

标准Hough变换用以检测直线特征需要较大的时间/空间消耗,难以适用于实时图像处理。提出了一种适用于DSP的快速直线检测方法,并研究了其在轨道特性分析中的应用。结合插值采样的方法构建低级图像,利用方向滤波滤除目标特征外的特征点,降低目标特征外直线的干扰。同时针对DSP结构采用了一种后向映射策略,直接对图像进行逐个像素的参数变换,减少了空间消耗。实验数据分析表明,可以大幅提高直线特征检测效率,大幅提高实时图像处理速度。

The standard Hough transformation needs large time/space consumption, which is unsuitable for real-time image processing,as well as its application in rail characterization. This paper proposes an improved line extraction algorithm. In low definition image generated by interpolation sampling, the useless points are filtered out by direction filter, in this way the interference of discrete noises is greatly eliminated. In view of the hardware structure of DSP, this paper propose an improved image processing strategy,which is more efficiently in pipelined structure. The experiment research and data analysis showe that the method can improve the algorithm efficiency,which is suitable for real-time image processing system.