ing at the drawbacks that for probabilistic Hough transform it consumes a lot of memory and the line endpoints searching is vulnerable to interference from reticular aggregation points, a probability-based local Hough transform optimization algorithm is proposed. The edge is classified into two categories: sortable and non-sortable. For the former, sampling points are randomly picked and combined with their adjacent points for straight line searching. For the latter, the local probabilistic Hough transformation is carried out in the region of interest which is established around the random edge point, the endpoints are searched after the line is detected and the slope is fixed in real time. The error lines resulting from mesh aggregation points are excluded by the interval counting and the total interval length limit method. Experiments were carried out by 500 images. The proposed algorithm consumes less than 1/3 of the time of the probabilistic Hough transform, and it is highly resistant to line mis-detection of meshed aggregation edge points. Line detection is more accurate and memory consumption is reduced by more than 90%, compared with the probabilistic Hough transform.