Abstract: A novel line detection method combining the Hough transform and an improved least squares method was proposed. Advantages and drawbacks of the Hough transform and the least squares method on line detection and detec- lion accuracy were analyzed. Robust and heuristics-free coarse detection of lines was realized by the Hough transform to determine image regions where a line may exist. Least squares regression was then applied on feature points in these re- gions to obtain accurate line parameters. To overcome the sensitivity of conventional least squares method to outliers, the xrleast square with dual removal algorithm, which deletes each iteration a pair of data points with maximum positive and negative fitting errors to ensure the reservation of normal points in the data set and thus guarantee the accuracy of the linear regression, was proposed. Experimental results show that the novel method gives higher detection rate and more accurate line parameters compared with the Hough transform Besides, lower Hough space resolution can be a- dopted without much impacts on the detection results, thus reducing the cost on the memory needed.

Key words: Line detection, Hough transform, Least squares