Automatic Road Sign Recognition From Video
3 likes3,758 views
The document describes an automatic road sign recognition system developed for high resolution roadside video. It uses a color space method to detect signs by recognizing color patterns. Specifically, it detects signs by converting RGB images to HSI color space and matching hue values. The system was tested on various sign types under different lighting conditions with good detection results. In conclusion, the approach provides a robust and fast way to detect most common road signs in New Zealand videos and identify the sign type and location.
Automatic Road Sign Recognition From Video
- 1. Automatic Road Sign Recognition from Video Dr Wei Liu Senior Engineer
- 2. Presentation Outline • Introduction • Methodology • Results • Conclusions
- 3. Introduction • Road signs provides important information for guiding, warning, or regulating the drivers’ behaviour in order to make driving safer and easier • The Road Sign Recognition (RSR) is a field of applied computer vision research concerned with the automatic detection and classification of traffic signs in traffic scene images acquired from a moving car.
- 4. Introduction • Pavement Management Services have developed the first (and currently only) truly spatially registered video system in Australia. • The digital video system offers continuous, high resolution video capture of five different views along the roadway.
- 5. Introduction • A road sign recognition system (RS2) has been developed for the high resolution roadside video recorded by PMS video system. • The recognition process of RS2 is divided into three distinct parts: • Detection and Location • Recognition and Classification • Display and record for information of road signs
- 6. Introduction • The PMS video system consists of five industrial quality digital cameras mounted in any directional configuration on a host vehicle. • The cameras works well in varying and low light conditions, at all times maintaining high shutter speeds to eliminate motion blur. • The cameras have individual image resolutions of: • 768x576 (broadcast quality of road asset views) • 1024x768 (high resolution image for pavement view)
- 7. Introduction • Image capture and survey position are determined by precision odometer and GPS location equipment. • The image capture trigger is accurate enough at synchronising the image captured to make panoramic views from collection of cameras at high test speed (100km/hr).
- 8. Introduction • Typically, high resolution images are collected for every one meter of the road surface and every ten meters of the roadside assets. • The spatial reference is achieved within the video itself by creating a ‘data-cloud’ of DGPS points for each frame of the video, which gives it the ability to locate and therefore ascribe a DGPS coordinates to any fixed item within the view of each of the five cameras.
- 9. Introduction • PMSVideo is a computer software tool used to enable the playback and examination of video collected using Pavement Management Services digital video system. • The PMSVideo software allows the user to find road sections according to the road owners road referencing scheme and even recording notes and other useful information for use in other road management systems.
- 10. Introduction • To ensure the creation of accurate location of road assets in the video, a grid calibration procedure for each camera is applied prior to the commencement of the survey. • After calibration, the PMS video system is able to provide a three dimensional plot from a two dimensional plot by mapping the world coordinate to the views presented by each camera with the same accuracy of DGPS data cloud.
- 11. Introduction • The difficulty in recognizing road signs is largely due to the following reasons: • The colors of road signs, particularly red, may fade after long exposure to the sun. • Air pollution and weather conditions may decrease the visibility of road signs. • Outdoor lighting conditions varying from day to night may affect the colors of road signs. • Obstacles, such as vehicles, pedestrians, and other road signs, may partially occulde road signs. • Video images of road signs will have motion blur if the camcorder is mounted on a moving vehicle due to vehicle vibration as well as motion.
- 12. Methodology • While lots of attempts at automated sign recognition were based on the detection of shape patterns, the proposed method for PMS Video detects road signs by recognising their patterns in color space.
- 13. Methodology • How can we quantitatively describe a color? • we usually treat colors as RGB triples. The three components define the amount of red, green, and blue, respectively, whose combination results in the desired color on a computer screen. Typically, each channel uses discrete values from 0 to 255. • The color space formed by all possible RGB values is also called the RGB space.
- 14. Methodology • The RGB color space is easy to use and represents color in the same way as the monitor requires it for its display. However, for computer vision applications such as the recognition of objects, other color spaces are more useful. • We will introduce the HSI color model, standing for hue, saturation, and intensity. • These dimensions characterize important object properties more naturally as compared to the RGB components.
- 15. Methodology • HSI Color Space • Hue is determined by the dominant wavelength in the spectral distribution of light wavelengths. • Saturation is the magnitude of the hue relative to other wavelengths. • It is defined as the amount of light at the dominant wavelength divided by the amount of light at all wavelengths. • Intensity is a measure of the overall amount of light within the visible spectrum. • It is a scale factor that is applied across the entire spectrum.
- 16. Methodology • HSI Color Space •Hue •Saturation •Brightness
- 17. Methodology • Conversion from RGB to HSI HSI RGB
- 18. Methodology • Conversion from RGB to HSI ) )( ( ) ( 2 2 arccos 2 B G B R G R B G R H − − + − − − = ) , , min( 3 1 B G R B G R S + + − = 3 B G R I + + =
- 19. Methodology • Advantages of using HSI color space for Sign Detection • It allows a better tolerance to changes in lighting conditions compared to other color models • A specific color can be recognized by matching a small range of hue value. • Ability to detect signs with different shape and detect composite signs
- 21. Results • Warning Signs-Normal Condition
- 22. Results • Warning Signs-Under shade
- 23. Results • Warning Signs-Under shade
- 25. Results • Warning Signs-uneven lamination
- 38. Results • Sign Recognition Results on Google Earth
- 39. Conclusions • An automatic road sign recognition module from road video collected by PMS video system was developed. • The proposed approach is robust and fast for detection of most of road signs commonly found in New Zealand, including warning signs, information signs, regulatory signs, and street signs. • The sign recognition results include the exact location , type of road sign occurred in the video frame, and the image containing the road signs detected, which can be used for road sign condition evaluation.