The invention, which relates to the field of the image processing technology, discloses a computer-vision-based detection method for a passenger flow volume. The detection method comprises: carrying out an OR operation on a grayscale value of each pixel point of a first foreground image and a grayscale value corresponding to each pixel point of a binary difference image to obtain a second foreground image; dividing the second foreground image into a plurality of third foreground images with same sizes and carrying out binaryzation processing on each third foreground image to obtain a fourth foreground image; extracting characteristic parameters in the fourth foreground image successively according to a sequence and determining whether a type of a moving target is a pedestrian type by operation by a BP neural network based on the characteristic parameters; and if so, extracting a coordinate position of the moving target; and comparing a horizontal coordinate of the moving target with a horizontal coordinate of a preset detection region to determine whether moving target exceeds a detection region. Therefore, the accuracy of passenger flow volume detection is improved.