The invention discloses a real-time constructor safety helmet detection method based on pedestrian detection. According to the method, firstly, large-scale marked samples are utilized to carry out multi-time iteration training for an Adaboost classifier extracted on the basis of ACF characteristics; secondly, the trained classifier is utilized to carry out pedestrian detection on a video sequence, and the precise position and the quantity of targets are determined; thirdly, locus analysis on a detection result is carried out, whether the detection result is wrong is determined according to the locus information, if a score of the detection result is not high, dump processing is carried out, the detection result with high score is tracked, a characteristic that a human head is at a top portion of the detection result is utilized, the portion of a safety helmet is locked, through color space conversion, and safety helmet detection is finally realized through utilizing a pixel ratio for determination according to the separation color channel statistics pixel information. The method is advantaged in that good robustness is realized for multiple occasions, and a pixel point statistics detection method based on HSV space has quite good effects for special color safety helmets.