Target rapid detection method based on ThunderNet

Abstract

The invention discloses a ThunderNet-based target rapid detection method. The method comprises the following steps of A1, preprocessing image information captured by a camera; A2, putting the capturedimage information into a Snet backbone network of ThunderNet to carry out target detection, and extracting a characteristic value; and A3, putting the data stream subjected to feature processing intoa detection part of ThunderNet, fitting the data by using two strategies of CEM and SAM, and performing supervised training through an RPN network to distinguish foreground features from background features so as to perform target identification. According to the method, the advantages of ThunderNet detection can be fully utilized, and information can be accurately and efficiently extracted fromthe captured image. Compared with a conventional single-stage detector, the method is higher in detection precision, and remarkably reduces the calculation cost. The method can be used for rapid and intelligent detection of correct wearing of the mask.

Description

technical field [0001] The invention relates to the fields of computer vision and digital image processing, in particular to a ThunderNet-based rapid target detection method. Background technique [0002] Face recognition is currently one of the most widely used image recognition applications, and its development has gone through several different stages. From early text recognition, to mid-term digital image processing and recognition, to current object recognition. Compared with simple text recognition, digital images are more efficient and powerful. They have huge advantages such as storage, convenient transmission and compression, less distortion during transmission, and convenient processing. These provide a strong impetus for the development of image recognition technology. Its license plate recognition in the security field has been widely used. The latest object recognition is the perception and understanding of objects and environments in the three-dimensional wor...

AI Technical Summary

Problems solved by technology

Mobile devices are computationally constrained compared to server-grade GPUs, imposing tighter constraints on the computational cost of detectors

Although face recognition technology is currently widely used, due to the epidemic, it is necessary to be able to efficiently identify targets wearing masks or even goggles, and the recognition efficiency is greatly reduced.

Moreover, current CNN-based detectors are resource-poor and require a large amount of computation to achieve the desired detection accuracy, which hinders their widespread use in mobile scenarios.

The current technical solution has the problems of slow calculation and processing speed and high processing cost

Images