基于直线生成的卷积霍夫线段检测OACSTPCD

As an upstream task in computer vision,line detection provides support for downstream tasks including industrial vision and remote sensing image analysis.One major direction in line detection is Hough line detection.However,the accumulators in existing Hough detection methods are generally designed based on approximate principles,leading to lower accuracy in line detection.To enhance the precision of Hough line detection,convolution is employed to improve the accumulator in Hough line detection and Convolutional Hough Line Segment Detection based on line generation(CHLSD)is proposed.Firstly,median filtering is utilized to smooth the complex textures in original images,and extract edges from smoothed images.Next,the extraction results are denoised using a bitwise Exclusive OR operation and candidate end points of line segments are retained.Then,Bresenham algorithm generates line segments with paired candidate end points,accumulator improved with convolution is used to determine if generated line segments match extracted edges.Finally,parameters for determined end points are calculated and lines with similar parameters are merged to obtain final detection result.Experiments show that CHLSD improves the detection precision while retaining the robustness of Hough line detection,achieving an F1 score of 0.762 6,which is superior to the comparison methods'score of 0.652 3.