Dynamic scene foreign matter intrusion detection method based on deep learning

Abstract

The invention discloses a dynamic scene foreign matter intrusion detection method based on deep learning. The method comprises the following steps: acquiring RGB image frames of a monitored scene by using monitoring equipment; segmenting a target region in the image by using the trained target segmentation model to obtain a target region binarization mask result; using the trained target detection model to detect the foreign object target in the image, and obtaining the category and position information of the foreign object target; calculating a pixel ratio P of a target area in the position boundary frame of the foreign object, and judging the invasion state of the foreign object according to the pixel ratio P; according to the category information of the foreign object target, determining the invasion level of the foreign object target; and finally, according to the invasion state and grade of the foreign object target, sending out corresponding safety early warning. Foreign matter intrusion detection can be completed only through a single frame of image, the intrusion state and the intrusion level of the foreign matter target are obtained at the same time, and an accurate and reliable basis is provided for safety forecasting and early warning of all activities in a dynamic scene.

Description

technical field [0001] The invention relates to the fields of digital image processing, pattern recognition and computer vision, and in particular to a deep learning-based foreign object intrusion detection method in a dynamic scene. Background technique [0002] In order to ensure the safety of people's daily production and life, some dangerous areas are usually artificially divided, and foreign object intrusion detection is carried out in this area. For example, the train track area, the mechanical processing area of ​​the production workshop, and some areas where idlers are prohibited from entering and exiting. These areas are usually relatively dangerous. If foreign objects invade, it will lead to activities that cannot be carried out normally, and even lead to personal safety accidents. Therefore, it is very necessary to improve the foreign object intrusion detection ability in these dangerous area scenarios, and to issue accurate and reliable safety warnings in the eve...

AI Technical Summary

Problems solved by technology

[0003]At present, foreign object intrusion detection technology is mainly based on traditional image processing methods such as background difference method or frame difference method, although these methods can detect moving objects in the current scene , but there are many restrictions

For example, the background difference method needs to establish a complex background model, and the parameters are difficult to adjust; the frame difference method is relatively simple, but it is very sensitive to noise, such as shadow changes and screen shakes, etc., which easily affect the detection results

Moreover, at present, the above two methods and the improved methods based on the above two methods are all for detection of the whole image area, and cannot be analyzed for a certain area in the picture

At the same time, it is worth mentioning that none of the above methods can classify the detected abnormal targets, and it is impossible to classify different hazard levels for different types of objects, which is prone to generate more false alarms, wasting manpower, material resources and financial resources

Images