Escalator retrogradation behavior detection method and device based on MobileNetV3 and optical flow algorithm

Abstract

The invention provides an escalator retrograde behavior detection method and device based on MobileNetV3 and an optical flow algorithm, and relates to the technical field of behavior detection. The method comprises the following steps: acquiring an escalator monitoring picture set, and carrying out image recombination and de-noising processing on the escalator monitoring picture set; the escalator monitoring picture set is input into a MobileNetV3 model, and the position of a person is obtained; acquiring a punctuation picture set through the figure position; inputting the punctuation picture set into an improved LK optical flow algorithm to obtain a route set; and comparing the route set, detecting a retrograde motion behavior and sending a prompt. According to the method, image enhancement is carried out on a monitoring video picture set through an adaptive non-local mean filtering method, and texture information of the image is better reserved while the image is smoothed; pyramid optimization is carried out on the LK optical flow algorithm, and a constraint function is added, so that the stability of tracking a target person on an escalator by the optical flow algorithm is improved; interference background factors contained in a monitoring image set are removed through a monitoring image recombination method, and the influence of a complex scene on the MobileNetV3 model is reduced.

Images