A novel method of local uncertainty based measure is proposed for line detection in the paper to tackle the problems of decrease in detection accuracy in noised images for multi-line detection of complicated work-pieces. The proposed scheme firstly partitions an object into several regions. Then a probability model of uncertainty that describes how an edge pixel belongs to a line is built within each region

and accumulated uncertainty measures for lines formed by any pair of two edge points are computed according to two point combination and the Bayesian rule. Lines are finally detected using soft voting in parameter spaces. The capability of anti-noise and fast processing speed is the key feature of the algorithm. Experimental results show that accuracy error of proposed method less than 1‰ when noise variance equals to 0.06 and detection accuracy can reach above 90%. The detection period is 1/2 of method of pure uncertainty measure and processing speed is 4~5 times faster than conventional hough transform (HT).