The invention discloses a text detection method based on a depth measurement model, comprising the following steps: step 1, obtaining a candidate region at a character level by using an MSER detectionalgorithm; 2, filtering the candidate region by use a classifier, and removing a non-character regions; 3, clustering the obtained characters into text lines according to the geometrical position information; 4, according to the heuristic rule, dividing the text line into concrete words; 5, constructing a word-level training set; 6, training a depth metric learning model; 7, classifying the textbox by using the depth measurement model obtained in the step 6, and obtaining the final text box area.