An Attention Network-Based Day/Night Image Classification and Object Detection Method

Abstract

The invention discloses a day / night image classification and object detection method based on an attention network. The method first collects street camera surveillance video and processes it into an image for labeling, and combines an open-source street data set to jointly construct an image data set; Pyramid’s deep convolutional neural network extracts the apparent features of the image; predicts the day / night attributes of the image on the extracted features, and captures attention maps that characterize day / night objects; weights the extracted feature maps based on the attention maps; finally According to the predicted day / night attribute, the weighted feature map is input into the detection head corresponding to day / night for position regression and object classification. The invention aims to make the network pay attention to the different features of day / night through the attention mechanism, and complete the detection of day / night objects respectively through two branches, which can improve the performance of day / night object detection, and can be used for street intelligent monitoring systems.

Description

technical field [0001] The invention belongs to the technical field of computer vision recognition, and in particular relates to an attention network-based day / night image classification and object detection method. Background technique [0002] Object detection is the basis of many other classic vision problems, and has great practical value and application prospects. It is an indispensable technical point in intelligent video surveillance, automatic driving, face recognition, robot navigation and other application fields. With the success of convolutional neural networks (CNNs), deep learning has proven to be an effective solution. [0003] Object detection needs to complete the following three tasks: [0004] 1. Distinguish between the foreground object frame and the background, and assign them appropriate category labels; [0005] 2. Regress a set of coefficients to maximize the intersection-over-union ratio (IoU) or other indicators between the detection frame and the...

AI Technical Summary

Problems solved by technology

However, the enhanced image produced by this technology often adds a lot of noise, which deviates from the feature distribution of the real image, which is not conducive to the detection performance of target detection.

In the application process, nighttime images need to pass through the enhanced network first, and then through the detection network. The time complexity is high and the detection speed is slow. It is not suitable for industrial application scenarios, such as a near-real-time intelligent video surveillance system.

In addition, model training is more complicated, and end-to-end training cannot be achieved.

Images