SPIE Startup Challenge 2015 Founding Partner - JENOPTIK Get updates from SPIE Newsroom
  • Newsroom Home
  • Astronomy
  • Biomedical Optics & Medical Imaging
  • Defense & Security
  • Electronic Imaging & Signal Processing
  • Illumination & Displays
  • Lasers & Sources
  • Micro/Nano Lithography
  • Nanotechnology
  • Optical Design & Engineering
  • Optoelectronics & Communications
  • Remote Sensing
  • Sensing & Measurement
  • Solar & Alternative Energy
  • Sign up for Newsroom E-Alerts
  • Information for:
    Advertisers
SPIE Defense + Commercial Sensing 2017 | Register Today

OPIE 2017

OPIC 2017

SPIE Journals OPEN ACCESS

SPIE PRESS

SPIE PRESS




Print PageEmail PageView PDF

Electronic Imaging & Signal Processing

A simple model for a lane detection system

An efficient algorithm for detecting lane markers uses the log-polar transform and random sample consensus to achieve robust performance under various road conditions.
15 February 2012, SPIE Newsroom. DOI: 10.1117/2.1201202.004086
A simple model for a lane detection system

Since the 1990s, lane detection algorithms have been an essential part of intelligent vehicle systems and have been used in many applications such as driver assistance, lane departure warning, and lane-keeping systems. Hence, this type of algorithm has received considerable attention. However, certain conditions are still a challenge: shadows, certain solar positions, climate, and occlusion.1 Many algorithms2–6 have been developed to achieve robustness against these difficult conditions. For example, Wang and colleagues developed the canny/Hough estimation of vanishing point method,5 whereas Aly proposed an approach that combines the plan-view and random sample consensus (RANSAC).6 Unfortunately, these algorithms still detect false lanes as a result of shadows from uniformly oriented tree trunks, telegraph poles, and buildings.

To solve these problems, we propose a simple, efficient lane marker detection system that uses a log-polar transform (LPT) and RANSAC.7 The system applies two key techniques. First, the new lane model based on an LPT converts Cartesian space coordinates (x, y) into a radius (log ρ) and angle (θ) relative to the origin of the coordinate system: see Figure 1(a). Regardless of whether it is straight or curved, a 2D lane is represented by parabolic curves in log-polar space: see Figure 1(b) and Figure 2(b). Second, we use the RANSAC fitting algorithm to fit the lane markers precisely: see Figure 2(c).


Figure 1. Lane modeling using the log-polar transform (LPT): (a) definition of LPT, and (b) the original image (left), the binary image with region of interest in red after pre-processing (center), and the image transformed by the LPT (right).

Figure 2. The results of the proposed lane detection algorithm: (a) binary images produced by pre-processing and (b) the corres-ponding LPT images. (c) The results of random sample consensus (RANSAC) parabolic fitting, and (d) the corresponding inverse LPT images.

Our proposed lane detection algorithm requires a human operator to provide an initial estimate of the road location. The operator selects the region of interest—see Figure 1(b): center—by setting parameters such as the coordinates of the upper-left corner of the region and its width and height. By contrast, other systems require an assumption regarding the road's structure (for example, that it is straight, with two parallel lanes) and then extract the vanishing point using the Hough transform; hence, the computational complexity is increased.

The algorithm begins by dividing the original image of the road into subregions. It then pre-processes each one by smooth filtering and adaptive thresholding: see Figure 2(a). This produces a candidate lane, to which an LPT and RANSAC curve fitting are applied. Specifically, a sample of the points available in the candidate lane is fitted with a uniform cubic B-spline using a least squares method. This step provides the control points for the spline that minimizes the sum of the squared error in fitting the sampled points. The algorithm uses three control points, which are sufficient for fitting the spline. As the first and third control points are in triplicate, the parabolic curve is actually computed from seven points.

During the experimental development of our algorithm, we collected a number of different images—from highways and urban streets, with and without shadows, straight and curved roads—using the lane detection system with a CCD camera (see Figure 3). The system was implemented in C++ using the open source OpenCV library and tested on real roads, including curved and straight roads. According to computer simulations, the proposed lane detection system exhibits lower computational complexity and better lane detection performance.


Figure 3. Our lane detection system with a CCD camera.

In summary, we have introduced a new lane model based on the LPT and a lane detection algorithm using RANSAC. To extract the optimal lane markers, the algorithm performs a pre-processing step and transforms the candidate lanes to log-polar space. Then the RANSAC B-spline fitting method is used to exactly match the lane markers. Computer simulations indicate that high lane detection performance can be achieved under various weather and road conditions. Our future work will focus on reducing false positive lane identifications and implementing a real-time system using a system on chip.

This work was supported by the Technology Innovation Program (Industrial Strategic Technology Development Program, 10033630) funded by the Ministry of Knowledge Economy, South Korea.


Ju-Young Kim
LED-IT Fusion Technology Research Center
Yeungnam University
Gyeongsan-si, South Korea

Ju-Young Kim received a his PhD from the School of Electrical Engineering and Computer Science, Kyungpook National University, South Korea, in 2009. He has worked as a senior researcher since 2010, focusing computer vision for intelligent vehicle technology and smart LED lighting.

Ja-Soon Jang
School of Electrical Engineering and Computer Science
and
LED-IT Fusion Technology Research Center Yeungnam University
Gyeongsan-si, South Korea

Ja-Soon Jang received his MS and PhD degrees from the School of Materials Science and Engineering, Gwangju Institute of Science and Technology (1998 and 2002, respectively). He is currently an associate professor and the chief director at the LED-IT Fusion Technology Research Center. His professional interests include gallium nitride-based LEDs, nanophotonics, and LED applications.


References:
1. C. R. Jung, C. R. Kelber, Lane following and lane departure using a linear-parabolic model, Image Vision Comput. 23, no. 13, pp. 1192-1202, 2005. doi:10.1016/j.imavis.2005.07.018
2. M. Chen, T. Jochem, D. Pomerleau, AURORA: a vision-based roadway departure warning system, Proc. IEEE Intell. Robots Syst. 1, pp. 243-248, 1995. doi:10.1109/IROS.1995.525803
3. A. Broggi, A massively parallel approach to real time vision based road marking detection, Proc. IEEE Intell. Veh. Symp., pp. 84-89, 1995. doi:10.1109/IVS.1995.528262
4. D. Pomerleau, Visibility estimation from a moving vehicle using the RALPH vision system, Proc. IEEE Intell. Transp. Syst., pp. 906-911, 1997. doi:10.1109/ITSC.1997.660594
5. Y. Wang, E. K. Teoh, D. Shen, Lane detection and tracking using B-Snake, Image Vision Comput. 22, pp. 269-280, 2004. doi:10.1016/j.imavis.2003.10.003
6. M. Aly, Real time detection of lane markers in urban streets, Proc. IEEE Intell. Veh. Symp., pp. 7-12, 2008. doi:10.1109/IVS.2008.4621152
7. J.-Y. Kim, H.-R. Lim, C.-S. Lee, J.-S. Jang, An efficient lane markers detection algorithm using log-polar transform and RANSAC, Proc. SPIE 8135, pp. 81351J, 2011. doi:10.1117/12.893529