YOLOv8n road multi-target detection method and system based on HGNet backbone network

Abstract

The invention discloses a YOLOv8n road multi-target detection method and a YOLOv8n road multi-target detection system based on an HGNet backbone network, and the method achieves the rapid and accurate extraction of road markers, such as lane markings and traffic signs, through the comprehensive analysis of multi-dimensional information, such as geometric features, coordinate information and reflection intensity, of point cloud data, and in combination with an intelligent extraction algorithm. An operator can complete intelligent tracking extraction of lane marking lines, arrows, zebra crossings and the like on a road with the length of hundreds of meters through simple clicking and basket pulling operation, and high-quality vector output is generated. According to the technology, manual intervention is remarkably reduced, the operation efficiency is improved, and important data support is provided for the fields of road asset management, intelligent traffic system construction and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent transportation, and particularly relates to a YOLOv8n road multi-object detection method and system based on the HGNet backbone network. Background Art

[0002] With the rapid development of intelligent transportation systems, the requirements for the accuracy and real-time performance of road information are increasing day by day. Traditional manual measurement and recognition methods are difficult to meet the needs of large-scale and high-efficiency extraction of road surface markers. Therefore, the automatic extraction technology based on point cloud data has emerged as the times require and has become one of the current research hotspots. There are problems of low efficiency and poor accuracy in the existing manual measurement and recognition methods.

[0003] HGNetv2 is a new generation of super backbone network released by Baidu in 2022, which has achieved state-of-the-art results on the ImageNet dataset. YOLOv8 is the next updated version of YOLOv5 open-sourced by the ultralytics company in 2023, and currently supports image classification, object detection, and instance segmentation tasks. yolov8 is a model family, including from small to large: yolov8n, yolov8s, yolov8m, yolov8l, yolov8x, etc. The parameters of yolov8n are as follows: size (pixels): 640, mAPval: 37.3, speed (CPU ONNX, ms): 80.4, speed (A100 TensorRT, ms): 0.99, parameters (M): 3.2, FLOPs (B): 8.7. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a YOLOv8n road multi-object detection method and system based on the HGNet backbone network to solve the deficiencies in the prior art.

[0005] To achieve the above object, the present invention is implemented by the following technical solutions:

[0006] On the one hand, a YOLOv8n road multi-object detection method based on the HGNet backbone network is provided, which includes:

[0007] S1. Organize the domestic and foreign open datasets CTSDB, CCTSDB, ApolloScape, TT 100K, GTSRB GTSDB, LISA, BSTLD BOSCH, and extract the dataset samples of traffic lights and road traffic signs;

[0008] S2. Use a laser scanner to obtain road point cloud data, and the road point cloud data includes geometric shapes, coordinate positions, and reflection intensity information;

[0009] S3. Perform preprocessing on the original point cloud data, including filtering, denoising, and registration;

[0010] S4. Based on the geometric features, coordinate information, and reflection intensity of the point cloud data, extract the potential features of traffic lights, lane markings, traffic signs, and road markers;

[0011] S5. Use an intelligent extraction algorithm to identify and classify the extracted features, achieving accurate identification of traffic lights, lane markings, arrows, and zebra crossing road markers;

[0012] S6. Convert the recognition result into a vector graphics format.

[0013] According to the YOLOv8n multi-object detection method for roads based on the HGNet backbone network, wherein the intelligent extraction algorithm combines deep learning and image processing technologies to achieve effective identification of traffic lights and markers in complex road environments.

[0014] According to the YOLOv8n multi-object detection method for roads based on the HGNet backbone network, wherein it further includes a human-computer interaction optimization step to optimize the human-computer interaction interface.

[0015] According to the YOLOv8n multi-object detection method for roads based on the HGNet backbone network, wherein it further includes a data processing acceleration step, and parallel processing and GPU acceleration technology means are used to improve the data processing speed.

[0016] On the other hand, a YOLOv8n multi-object detection system for roads based on the HGNet backbone network is provided, which is implemented based on the YOLOv8n multi-object detection method for roads based on the HGNet backbone network as described in any one of the above, and includes:

[0017] A data acquisition module for obtaining road point cloud data;

[0018] A preprocessing module for performing preprocessing on the original point cloud data, including filtering, denoising, and registration;

[0019] A feature extraction module for extracting the potential features of traffic lights, lane markings, traffic signs, and road markers based on the geometric features, coordinate information, and reflection intensity of the point cloud data;

[0020] An intelligent recognition module for using an intelligent extraction algorithm to identify and classify the extracted features, achieving accurate identification of traffic lights, lane markings, arrows, and zebra crossing road markers;

[0021] A mapping output module for converting the recognition result into a vector graphics format.

[0022] The beneficial effects of the technical solution of the present invention are:

[0023] By comprehensively analyzing multi-dimensional information such as the geometric features, coordinates, and reflection intensity of the point cloud, and combining intelligent extraction algorithms and deep learning models, the present invention realizes the rapid and accurate extraction and vectorization of road markers such as traffic lights, lane markings, and traffic signs. This technology significantly reduces manual intervention and improves operation efficiency, providing important data support for fields such as road asset management and the construction of intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] To further illustrate the above-mentioned objects, structural features, and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Figure 1 It is the overall network structure diagram of a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The terms "invention" and "the present invention" as used in this specification are intended in a broad sense to refer to all subject matter of this specification and any following patent claims. Statements containing these terms should not be construed as limiting the subject matter described herein or the meaning or scope of any following patent claims. Additionally, this specification does not attempt to describe or limit the subject matter covered by any specific component, paragraph, statement, or claim of this application's drawings. The subject matter should be understood with reference to the entire specification, all drawings, and any following claims. The present invention may have other embodiments and may be practiced or implemented in other ways. Moreover, it should be understood that the wording and terms used herein are for illustrative purposes and should not be considered limiting.

[0027] Details of the present invention will now be discussed with reference to the drawings of the present invention, which are given by way of example only. In the drawings, like features or components may be labeled with the same reference numerals.

[0028] The use of the terms "comprising", "having", and "including" and their variants herein means including the items listed hereinafter, their equivalents, and additional items. Although directions such as above, below, upward, downward, backward, bottom, top, front, and back may be referred to in the description of the drawings for convenience with reference to the drawings, these directions are not intended to be literally accepted or limit the present invention in any form. Additionally, terms such as "first", "second", "third", etc. are used herein for illustrative purposes and are not intended to indicate or imply importance or significance.

[0029] Refer to Figure 1 As shown, the YOLOv8n road multi-object detection method and system based on the HGNet backbone network of the present invention includes:

[0030] Step S1: Organize the open datasets at home and abroad, namely Chinese Traffic Sign Database (CTSDB), the dataset CCTSDB obtained by Changsha University of Science and Technology after expanding on the basis of the CTSDB dataset, Baidu ApolloScape dataset, Tsinghua-Tecent 100K Tutorial (TT 100K), GTSRB GTSDB (Germany), LISA (USA), BSTLDBOSCH (Bosch, Germany), and extract the dataset samples of signal lights and road traffic signs. (Dataset organization module)

[0031] Step S2: The data acquisition module uses a laser scanner to obtain road point cloud data, and the point cloud data contains information such as geometric shape, coordinate position, and reflection intensity.

[0032] Step S3: The preprocessing module performs preprocessing such as filtering, denoising, and registration on the original point cloud data to improve the data quality. Commonly used preprocessing methods include mean filtering, median filtering, and Gaussian filtering, etc., as well as a point cloud clustering algorithm based on K-means clustering to extract road point cloud data.

[0033] Step S4: Based on the geometric features of the point cloud (such as point cloud density, slope), coordinate information, and reflection intensity, the feature extraction module extracts feature points related to road markers. For example, the noise is removed and the road edge line is extracted through a point cloud fitting algorithm, or key information such as lane lines and traffic signs is obtained using a feature extraction algorithm.

[0034] Step S5: The intelligent recognition module uses intelligent algorithms such as deep learning and template matching to recognize the extracted features. For example, a convolutional neural network (CNN) is used to process the imageized point cloud data to recognize markers such as signal lights, lane markings, and arrows. At the same time, the type and position of the markers are further confirmed through template matching technology.

[0035] Step S6: The mapping output module converts the extracted point cloud data into vector graphics for subsequent applications such as map drawing and data analysis. The vectorization process includes steps such as serialization of point cloud data, coordinate transformation, and graphic generation.

[0036] Through comprehensive analysis of multi-dimensional information such as the geometric features, coordinates, and reflection intensity of the point cloud, combined with intelligent extraction algorithms and deep learning models, the present invention realizes the rapid and accurate extraction and vectorization of road markers such as signal lights, lane markings, and traffic signs. This technology significantly reduces manual intervention, improves the operation efficiency, and provides important data support for fields such as road asset management and the construction of intelligent transportation systems.

[0037] The above are only the preferred embodiments of the present invention, and do not limit the implementation manners and protection scope of the present invention. For those skilled in the art, it should be realized that all the solutions obtained by equivalent substitution and obvious changes made by using the specification and illustrated content of the present invention should be included in the protection scope of the present invention.

Claims

1. A YOLOv8n multi-object detection method for roads based on the HGNet backbone network, characterized in that, Including: S1. Sort out the domestic and foreign open dataset CTSDB, CCTSDB, ApolloScape, TT 100K, GTSRB GTSDB, LISA, BSTLD BOSCH, and extract the sample datasets of traffic lights and road traffic signs; S2. Use a laser scanner to obtain road point cloud data, which includes geometric shapes, coordinate positions, and reflection intensity information; S3. Perform preprocessing on the original point cloud data, including filtering, denoising, and registration; S4. Based on the geometric features, coordinate information, and reflection intensity of the point cloud data, extract the potential features of traffic lights, lane markings, and traffic sign road markers; S5. Apply an intelligent extraction algorithm to identify and classify the extracted features, and achieve accurate identification of traffic lights, lane markings, arrows, and zebra crossing road markers; S6. Convert the recognition results into a vector graphics format.

2. The YOLOv8n multi-object road detection method based on the HGNet backbone network according to claim 1, characterized in that, The intelligent extraction algorithm combines deep learning and image processing technologies to achieve effective identification of traffic lights and markers in complex road environments.

3. The YOLOv8n road multi-object detection method based on the HGNet backbone network according to claim 1, characterized in that, It also includes a human-computer interaction optimization step to optimize the human-computer interaction interface.

4. The YOLOv8n road multi-object detection method based on the HGNet backbone network according to claim 1, characterized in that, It also includes a data processing acceleration step, which uses parallel processing and GPU acceleration technology means to improve the data processing speed.

5. A YOLOv8n road multi-object detection system based on the HGNet backbone network, characterized in that, Implemented based on the YOLOv8n road multi-object detection method based on the HGNet backbone network as described in any one of claims 1 to 4, including: A data acquisition module to obtain road point cloud data; A preprocessing module to perform preprocessing on the original point cloud data, including filtering, denoising, and registration; A feature extraction module to extract the potential features of traffic lights, lane markings, and traffic sign road markers based on the geometric features, coordinate information, and reflection intensity of the point cloud data; An intelligent recognition module to apply an intelligent extraction algorithm to identify and classify the extracted features, and achieve accurate identification of traffic lights, lane markings, arrows, and zebra crossing road markers; A mapping output module to convert the recognition results into a vector graphics format.

Images