Target detection method of YOLO algorithm based on batch re-normalization processing

Abstract

The invention relates to a target detection method of a YOLO algorithm based on batch re-normalization processing, and belongs to the technical field of target detection and computer vision. The target detection method comprises the following steps: inputting an image with the size of 448 * 448 into a network structure detection model based on a YOLO algorithm; segmenting the image into S * S grids, wherein if it is judged that the center of the object falls on a certain grid, the grid is responsible for detecting the object; predicting the posterior probability that the detected object in charge of the grid belongs to a specific class in the C classes; predicting B target frames and confidence scores for each personal grid; and calculating the correlation confidence of the target frame class. According to the invention, the regional suggestion type target detection framework is reformed, and repeated work caused by overlapping of suggestion boxes is avoided, and the target detection speed is improved.

Description

technical field [0001] The invention relates to a target detection method based on the YOLO algorithm of batch renormalization processing, and belongs to the technical fields of target detection and computer vision. Background technique [0002] Target detection is to find out the correct position of the target and determine the category of these targets, which is one of the core problems in the field of computer vision. Object detection has always been one of the most challenging problems in the field of computer vision due to the different appearance, shape, and posture of various objects, coupled with the interference of factors such as illumination and occlusion when imaging. In addition to image classification, target detection also needs to solve the core problems that a) the target may appear in any position of the image; b) the target has various sizes; c) the target may have various shapes. Although object detection has developed rapidly in recent years and achieve...

AI Technical Summary

Problems solved by technology

Although the R-CNN algorithm has achieved a 50% performance improvement compared with the traditional target detection algorithm, it also has defects: for the target detection method based on region proposals, the R-CNN series all need to generate a suggestion frame, and perform Classification and regression, but there is overlap between the suggestion boxes, which will bring a lot of repetitive work; the training test is not concise, the candidate region extraction, feature extraction, classification, and regression are all operated separately, and the intermediate data needs to be saved separately; the speed is slow ;The input image Patch must be forcibly scaled to a fixed size, which will cause object deformation and lead to a decrease in detection performance

Images