CN-121982678-A - Device and method for identifying traffic sign around traveling vehicle in low-illumination environment

CN121982678ACN 121982678 ACN121982678 ACN 121982678ACN-121982678-A

Abstract

The invention relates to the technical field of intelligent traffic, in particular to a device and a method for identifying traffic marks around a running vehicle in a low-light environment, comprising the following steps of acquiring original image data of a surrounding scene when the vehicle runs in the low-light environment; the method comprises the steps of image data joint optimization, traffic sign feature detection, identification result optimization and output, wherein the step of image data joint optimization comprises the steps of carrying out joint optimization processing comprising noise removal and feature enhancement on an original image data sequence to obtain optimized image data, the step of traffic sign feature detection comprises the steps of inputting the optimized image data into a traffic sign detection model based on an improved YOLO architecture to obtain preliminary detection data of traffic signs, the step of carrying out time sequence consistency check on the preliminary detection data and outputting stable traffic sign identification result data, and the method improves the robustness and the practicability of the whole system in a dynamic driving environment.

Inventors

LI DAHAI
CHEN SU
ZHANG YUHUA
ZHU XIAOHUI
MA MINGHUI
QI RENLONG

Assignees

郑州科技学院

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The method for identifying the traffic sign around the running vehicle in the low-light environment is characterized by comprising the following steps of: the original image data acquisition step is to acquire an original image data sequence of a surrounding scene when the vehicle runs in a low-light environment; the image data joint optimization step is that joint optimization processing comprising noise removal and feature enhancement is carried out on the original image data sequence to obtain optimized image data; Inputting the optimized image data into a traffic sign detection model based on an improved YOLO architecture to obtain preliminary detection data of traffic signs, wherein the improvement comprises the steps of introducing a channel and a spatial attention mechanism into the YOLO architecture and performing multi-scale feature fusion; And optimizing and outputting the identification result, namely performing time sequence consistency check on the preliminary detection data and outputting stable traffic sign identification result data.
2. The method for identifying traffic signs around a traveling vehicle in a low-light environment according to claim 1, wherein the image data joint optimization step specifically comprises: the denoising intermediate data generation sub-step is that the original image data sequence is divided into areas according to the pixel brightness statistical characteristics of the original image data, and a filtering algorithm adapted to the characteristics of each area is adopted for processing, so that denoised image data is generated; And a characteristic enhancer step, namely processing the denoised image data by adopting an image enhancement algorithm based on a Retinex theory, decomposing the denoised image data into reflection component data and illumination component data, and respectively optimizing the reflection component data and the illumination component data to generate the optimized image data.
3. The method for identifying traffic signs around a traveling vehicle in a low-light environment according to claim 2, wherein in the step of generating denoising intermediate data, the area division is performed according to the statistical characteristics of pixel brightness, specifically: Calculating a brightness segmentation threshold based on pixel brightness statistical features of the original image data; For each pixel in the raw image data, comparing its luminance value to the luminance segmentation threshold: if the brightness value of the pixel is lower than the brightness segmentation threshold value, classifying the pixel into a dark area, and adopting a first type image filtering algorithm for processing; If the brightness value of the pixel is higher than or equal to the brightness segmentation threshold value, classifying the pixel into a non-dark area, and adopting a second type image filtering algorithm for processing; wherein the smoothness of the first type image filtering algorithm is stronger than that of the second type image filtering algorithm.
4. The method for identifying traffic signs around a traveling vehicle in a low-light environment according to claim 1, wherein in the traffic sign feature detection step, the backbone network of the improved YOLO architecture performs feature extraction on the optimized image data to generate a feature map; in the feature extraction process, the channel and the spatial attention mechanism are applied to carry out self-adaptive weighting on the channel dimension and the spatial dimension of the feature map; and after the backbone network completes the spatial pyramid pooling operation, executing the multi-scale feature fusion operation to fuse the feature information from different sensing fields.
5. The method for identifying traffic sign around a traveling vehicle in a low-light environment according to claim 1, wherein in the traffic sign feature detection step, the YOLO architecture uses anchor frame parameters generated based on low-light traffic sign sample size cluster analysis, and performs bounding box regression optimization by using a Wise-IoU loss function.
6. The method for identifying traffic signs around a traveling vehicle in a low-light environment according to claim 1, wherein the step of optimizing and outputting the identification result includes performing smoothing processing on position information of the same traffic sign identified in consecutive multi-frame images by using a time sequence filtering algorithm, and eliminating abnormal identification results in which the position change between frames exceeds a preset distance threshold.
7. The method for identifying traffic signs around a traveling vehicle in a low-light environment according to claim 6, wherein the time-series filtering algorithm is a kalman filtering algorithm.
8. The method of claim 1, wherein the traffic sign recognition result data includes a traffic sign category, a confidence level, an estimated distance to the vehicle, and an azimuth angle.
9. A traffic sign recognition device for a traveling vehicle in a low-light environment, comprising: The image acquisition module is used for acquiring an original image data sequence of a surrounding scene when the vehicle runs in a low-light environment; The image joint optimization module is in communication connection with the image acquisition module and is used for carrying out joint optimization processing comprising noise removal and feature enhancement on the original image data sequence to obtain optimized image data; The traffic sign detection module is in communication connection with the image joint optimization module and is used for running a traffic sign detection model based on an improved YOLO architecture so as to process the optimized image data to obtain preliminary detection data of traffic signs, wherein the improvement comprises the introduction of a channel and space attention mechanism and a multi-scale feature fusion unit in the YOLO architecture; the recognition result optimizing and outputting module is in communication connection with the traffic sign detecting module and is used for carrying out time sequence consistency check on the preliminary detection data and outputting stable traffic sign recognition result data.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Description

Device and method for identifying traffic sign around traveling vehicle in low-illumination environment Technical Field The invention relates to the technical field of intelligent traffic, in particular to a device and a method for identifying traffic signs around a running vehicle in a low-light environment. Background In intelligent driving systems, reliable identification of traffic signs is critical to driving safety. However, in low-illumination scenes such as night, tunnels or overcast and rainy, the problems of uneven brightness, enhanced noise, blurred characteristics and the like of the images obtained by the vehicle-mounted camera often exist, and the performance of the traditional recognition system is severely restricted. In the prior art, to improve the usability of the low-illumination image, a two-stage processing flow of "first enhancement and then detection" is generally adopted. For example, chinese patent application No. cn202211187053.X discloses a method for detecting and identifying traffic signs in low-illuminance environments, which constructs an enhanced network based on Retinex theory, optimizes low-illuminance image quality by introducing an attention mechanism, and then uses a detection model to identify traffic signs. The application has the following limitations that the steps of enhancing and detecting are sequentially executed, accumulated processing delay is larger, the real-time requirement of a high-speed driving scene cannot be met, the network structure parameter is large, the operation efficiency is low on a vehicle-mounted platform with limited calculation force, the enhancement network takes the improvement of visual appearance as a main target, the output of the enhancement network does not meet the optimal feature extraction requirement of the detection network, an effective denoising mechanism is not designed at the enhancement front end, and noise can be further amplified in the enhancement process to interfere with subsequent detection. Disclosure of Invention In order to solve the technical problems, the invention provides a device and a method for identifying traffic signs around a driving vehicle in a low-light environment. The invention discloses a method for identifying traffic signs around a running vehicle in a low-light environment, which comprises the following steps: the original image data acquisition step is to acquire an original image data sequence of a surrounding scene when the vehicle runs in a low-light environment; the image data joint optimization step is that joint optimization processing comprising noise removal and feature enhancement is carried out on the original image data sequence to obtain optimized image data; Inputting the optimized image data into a traffic sign detection model based on an improved YOLO architecture to obtain preliminary detection data of traffic signs, wherein the improvement comprises the steps of introducing a channel and a spatial attention mechanism into the YOLO architecture and performing multi-scale feature fusion; And optimizing and outputting the identification result, namely performing time sequence consistency check on the preliminary detection data and outputting stable traffic sign identification result data. Preferably, the image data joint optimization step specifically includes: the denoising intermediate data generation sub-step is that the original image data sequence is divided into areas according to the pixel brightness statistical characteristics of the original image data, and a filtering algorithm adapted to the characteristics of each area is adopted for processing, so that denoised image data is generated; And a characteristic enhancer step, namely processing the denoised image data by adopting an image enhancement algorithm based on a Retinex theory, decomposing the denoised image data into reflection component data and illumination component data, and respectively optimizing the reflection component data and the illumination component data to generate the optimized image data. Preferably, in the denoising intermediate data generating sub-step, the area division is performed according to a pixel brightness statistical feature, specifically: Calculating a brightness segmentation threshold based on pixel brightness statistical features of the original image data; For each pixel in the raw image data, comparing its luminance value to the luminance segmentation threshold: if the brightness value of the pixel is lower than the brightness segmentation threshold value, classifying the pixel into a dark area, and adopting a first type image filtering algorithm for processing; If the brightness value of the pixel is higher than or equal to the brightness segmentation threshold value, classifying the pixel into a non-dark area, and adopting a second type image filtering algorithm for processing; wherein the smoothness of the first type image filtering algorithm is stronger than that of the second t