Random garbage throwing behavior detection method based on skeleton point fusion cyclic hole convolution

Abstract

The invention discloses a random garbage throwing behavior detection method based on skeleton point fusion cyclic hole convolution. The method comprises the following steps: collecting an image set containing littering behaviors for pre-training; obtaining an image training set of littering behavior individuals in the pre-trained image set, and artificially defining human body skeleton point distribution of littering behaviors for images in the image training set; making a skeleton point heat map of each image in the image training set based on human skeleton point distribution; constructing alittering behavior detection network based on skeleton point fusion cyclic hole convolution; inputting the pre-trained image set into a littering behavior detection network, and iteratively updatingthe network by using a gradient descent method to obtain an optimal littering behavior detection network; and inputting a plurality of continuous frames of detection images in the test set into the optimal littering behavior detection network, obtaining a corresponding skeleton point distribution sequence, carrying out similarity calculation, and judging whether a littering behavior exists or not.According to the invention, the littering behavior can be accurately detected in a complex scene.

Description

technical field [0001] The invention relates to computer vision and image recognition, in particular to a method for detecting littering behavior based on skeletal point fusion loop hole convolution, and belongs to the technical field of computer vision image processing. Background technique [0002] At present, the country is in the stage of rapid urbanization, and the environmental problem is an urgent problem to be solved. Using the current image processing technology to detect littering behavior is of great significance for building a beautiful town. There is already a throwing recognition method based on image semantic information. It is based on trajectory detection. By drawing the trajectory of the thrown object and the background boundary, it can be judged whether it is littering. However, this method relies heavily on the image background and its application range is narrow. Due to the mature application of deep learning in image recognition, a method for detecting...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to address the deficiencies in the prior art, to provide a littering behavior detection method based on skeletal point fusion cycle hole convolution, which can adapt to littering behavior detection in complex environments, to solve the current problem based on human body posture detection. The method is difficult to accurately estimate the bone points of the characters in the video under occlusion and complex environments, and the problem of large amount of calculation

Images