Machine vision detection method

The invention discloses a machine vision detection method. The machine vision detection method comprises the following steps: step 1, obtaining an image; step 2, image preprocessing, including preprocessing the image, extracting three channels from the original RGB image, dividing the obtained three channels into three gray level images, and respectively carrying out subsequent feature extraction;step 3, TLBP feature extraction, including the grayscale images of the three channels acquired in the step 2 are respectively subjected to the following processing: (1) partitioning the grayscale images; (2) solving a TLBP feature histogram of each partition; according to the method. Firstly, the TLBP feature is obtained by changing a neighborhood mean value as a threshold value extracted from the LBP feature. Secondly, the CTLBP feature is obtained by adopting the cascading feature of the RGB three-channel image, and the stability of feature extraction and the precision of pedestrian detection are improved through t