A fully convolutional examination room target detection method based on cascaded attention and point supervision mechanism

Abstract

A full convolution test target detection method based on cascading attention and point supervision mechanism, which consists of building a full convolution test target detection network model, training the full convolution test target detection network model, and detecting test set images. On the basis of the FCOS method, the present invention proposes a cascade attention module, which is inserted between the backbone network and the feature pyramid network, so that the feature pyramid network can obtain significant features; a point supervision branch is added to the full convolution detection head module, The detection model has the ability to distinguish overlapping targets, and can detect overlapping targets in the monitoring scene of the examination room with dense seats of candidates, and solves the problem of inaccurate regression of target detection bounding boxes in dense scenes. The accuracy of the invention is 92.9%, and the detection speed is 22.1 frames per second, which improves the accuracy and speed of the object detection task in the examination room, and can be used for detection in the examination room.

Description

technical field [0001] The invention belongs to the cross-research technical field of educational science and image processing, and in particular relates to examinee target detection in a standardized examination room. Background technique [0002] Examination is the main way to assess the knowledge level and ability of learners. In order to regulate the order of the examination room, prevent candidates from cheating, and maintain the fairness and impartiality of the examination, our country currently mainly uses the combination of on-site examiner inspection and electronic video surveillance to conduct examinations. Due to the large amount of data and redundant information in the examination room monitoring video, the efficiency of the traditional examination room monitoring system is extremely low, and the monitoring personnel often suffer from visual fatigue due to high labor intensity. Abnormal behavior of candidates. Using advanced computer vision technology to serve ...

AI Technical Summary

Problems solved by technology

[0005] Considering the relatively dense distribution of examinee targets in the monitoring video of the real examination room and the large difference in imaging size due to the distance between the sitting position and the imaging device, it is difficult for the FCOS method to accurately detect them

In addition, in the FCOS method, if the marked real frames overlap, and the position (x, y) is mapped to the original image, multiple real frames will be obtained, then this position is considered to be a fuzzy sample, and the overlapping real frames may be lost in the training process. create intractable ambiguity

In the classroom monitoring scene, the targets are densely distributed, and most of the targets will occlude each other. It is difficult to determine the boundary of the detection frame only by using the multi-level prediction and Center-ness method in the FCOS method, which is easy to cause misjudgment.

Images