Regarding the task of lane line detection, we need to clarify several questions:
1. What is the representation form of lane lines Output Type: Mask/Point Set/Vector Line Instantiation: Does each lane line form an instance Classification: Has lane lines been classified (single white, double yellow, etc.) Predefined Parameters: Can only detect a fixed number of lane lines Lane markings: Has lane markings also been detected
This will affect the annotation of data and the output form of the network, And what is ultimately needed is the equation of the lane line in the world coordinate system. Neural networks are more suitable for extracting features at the image level, and direct regression equation parameters have many limitations. So, the network inference output requires relatively complex post-processing to solve the problem of real coordinates
Note 1: Attached at the end of the article is the SLAM communication group
Note 2: Summary of Computer Vision Books
Note 3: It's not easy to organize, please click like
2. Where does the neural network inference go?
Image segmentation schemes tend to classify pixels, determine whether they belong to lane markings, and determine the classification of the markings; Image detection schemes tend to abstract segmentation boundary lines, determine the presence of lane lines in a set of anchors, and regress the position parameters of lane lines< p> The pixel by pixel lane markings themselves or represent abstract lane separation boundary linesThe dataset
requires a balance of data across various scene categories, such as highways, auxiliary roads, winding mountain roads, night time, rainy days, etc. Data checks should be conducted to filter out high-quality images. For example, for images with blurry and similar nighttime data on highways and videos of driving in the rain, frame annotations can be taken, and one image can be extracted every 10. For example, if multiple images at low speeds are similar, the accuracy will be falsely high and the accuracy of small categories will be increased, For example, viewing the histogram of lane line coefficients, and then rotating it slightly to make each coefficient distribution more reasonable and normalize the data, accelerating convergenceFunction of testing report:
1. Project bidding: Issue authoritative third-party CMA/CNAS qualification report
2. Online e-commerce platform entry: Quality inspection report recognized by major e-commerce platforms 3. Used as a sales report: issuing legally effective testing reports to make consumers more confident 4. Papers and research: Provide professional personalized testing needs 5. Judicial services: providing scientific, fair, and accurate testing data 6. Industrial problem diagnosis: Verify the troubleshooting and correction of industrial production problemsBaijian and testing process:
1. Telephone communication and confirmation of requirements
2. Recommend solutions and confirm quotations 3. Mail samples and arrange testing 4. Progress tracking and result feedback 5. Provide reports and after-sales service 6. If urgent or priority processing is requiredTesting and testing characteristics:
1. The testing industry is fully covered, meeting different testing needs
2. Fully cover the laboratory and allocate localized testing nearby3. Engineers provide one-on-one services to make testing more accurate
4. Free initial testing, with no testing fees charged
5. Self service order delivery for free on-site sampling
6. Short cycle, low cost, and attentive service 7. Possess authoritative qualifications such as CMA, CNAS, CAL, etc 8. The testing report is authoritative and effective, and is generally used in China