Two-wheeled vehicle helmet detection method based on improved FCOS and embedded grouping

Abstract

The invention discloses a two-wheeled vehicle helmet detection method based on improved FCOS and embedded grouping, and the method comprises the following steps: S1, obtaining image data of a two-wheeled vehicle road, and making a two-wheeled vehicle helmet detection data set; s2, preprocessing a two-wheeled vehicle helmet detection data set; s3, constructing a target detection model based on improved FCOS and association grouping; and S4, optimizing the constructed target detection model based on the improved FCOS and the association grouping by using a stochastic gradient descent method. The improved FCOS target detection algorithm is used for pixel-level prediction, model training parameters are greatly reduced, the requirements of high precision and real-time performance are met, a detection task and a grouping task can be integrated in one network, and the detection efficiency is improved. Member distribution differences between the prediction group and the real group and between different prediction groups are measured through cosine similarity, and the two-wheeled vehicle and the rider are accurately matched.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a two-wheeled vehicle helmet detection method based on improved FCOS and embedded grouping. Background technique [0002] In recent years, the use of two-wheeled vehicles represented by motorcycles and electric bicycles has increased rapidly in China. At the same time, due to the lack of safety protection awareness of riders, traffic accidents of two-wheeled vehicles have occurred frequently, and the number of deaths has increased year by year. Relevant studies have shown that the correct wearing of safety helmets can reduce the risk of traffic accident death by 60% to 70%, which plays an important role in protecting the life safety of riders. Therefore, the nationwide deployment of the "One Helmet, One Belt, One Belt" safety protection operation is carried out to investigate and correct the behavior of motorcycle and electric bicycle riders who do not wear safety helmets...

AI Technical Summary

Problems solved by technology

However, the weight allocation strategy called centrality assigns smaller weights to small targets with a small number of positive samples, reducing the contribution in the loss; large targets have more positive samples, assigning larger weights, resulting in redundant losses Contribution, which in turn affects the effect of the final target detection

Images