The invention discloses a method and device for detecting
dirt of a binocular ADAS camera. The
dirt can be efficiently detected. The method comprises the steps that A, whether a vehicle is in a movingstate or not is judged, if yes, the step B is executed, and if not,
processing is not performed; B, judging whether the number of motion frames is greater than a threshold value 1 or not, if so, considering that the current vehicle is in motion, carrying out the step C, and if not, do not carrying out
processing; C, whether the images of the two cameras are relatively consistent is judged, if yes, no
dirt exists, and if not, the step D is executed, specifically, the step C comprises the steps that C1, the
structural similarity is judged, if yes, the step D is executed, and otherwise, the stepC2 is executed; C2, gray scale comparison is carried out, if it is judged that dirt exists through gray scale comparison, the step D is carried out, and otherwise it is judged that no dirt exists; and D, judging whether the frame number of the dirty image is greater than a threshold value 2, if so, judging that dirt exists, and if not, judging that no dirt exists.