Discipline: Technology and Engineering
Subcategory: Electrical Engineering
Joshua Peeples - University of Alabama at Birmingham
Co-Author(s): Mohammed Al-Qizwini and Hayder Radha, Michigan State University, East Lansing, MI
Each year millions of casualties and fatalities occur as a result of car accidents [1]. A majority of these car accidents are due to driver faults such as drunken driving, texting while driving, driving while tired, negligent driving and other preventable acts [2]. The prevalence of preventable accidents has been a key motivation for the advent of self-driving vehicles, which are anticipated to mitigate vehicle-related fatalities and injuries significantly. Several aspects of autonomous vehicles that are necessary include lane, road landmark, car, pedestrian and sign detections in order to provide the autonomous vehicle with all the required information to make the correct decisions and navigate safely to its destination; several algorithms have been introduced in order to satisfy these components of autonomous vehicles. In 2008, Dr. Mohamed Aly published a landmark discovery method [3]- his algorithm for lane detection has been referenced or used by several other colleagues and professionals in this field. This study will document another attempt to further improve Dr. Aly’s original procedure for lane detection by first, transforming the images to the YCbCr model and using the intensity channel (Y) for the edge detection process which is expected to provide better contrast than the red channel in the RGB model that was used previously, thus facilitating the edge detection process of the lane marks. Secondly, the work of Dr. Aly will be extended to be compatible with different datasets besides the original dataset utilized by proportionally altering the camera parameters based on the image resolution. Finally, a road sign and vehicle detection feature that employs the Computer Vision technique of blob detection will be incorporated into the package in order to add another important function towards the autonomous car project. The use of the intensity channel proved effective in improving the accuracy while not accruing a significant amount of additional processing time. The improvement in the correct detection rate was most notable for the one of the four original datasets where the percentage of correct detections improved from about 87% to 95%. Secondly, the incorporation of datasets was demonstrated to be plausible without knowing the explicit camera settings as the algorithm was able to detect lanes in various images. Lastly, the traffic sign and vehicle detection was implemented successfully in addition to the lane detection feature in the original and additional datasets. Future work consists of adding tracking to eliminate some of the false positives acquired during the detection process and to also generate statistical data for the other datasets utilized during this study by acquiring ground truths for the number of lanes, road signs, and vehicles.
References: [1] ‘Road Crash Statistics’, Asirt.org, 2016. [Online]. Available: http://asirt.org/initiatives/informing-road-users/road-safety-facts/road-crash-statistics. [Accessed: 17- Jun- 2016].
[2] S. Singh, ‘Critical Reasons for Crashes Investigated in the National Motor Vehicle Crash Causation Survey’, 1st ed. Washington, DC: NHTSA’s National Center for Statistics and Analysis, 2015, pp. 1-2.
[3] M. Aly, ‘Real time detection of lane markers in urban streets’, Intelligent Vehicles Symposium, 2008 IEEE, June 2008, pp. 7-12.
Funder Acknowledgement(s): I thank Hayder Radha and Mohammed Al-Qizwini of the Wireless and Video Communications (WAVES) lab for assistance with the project. Funding was provided through Michigan State University's BEACON Summer Research Opportunities Program (SROP).
Faculty Advisor: Hayder Radha, radha@egr.msu.edu
Role: I primarily implemented our ideas into the algorithm as well as researched and develop methods to achieve our aims. For the use of the new color model, I needed to determine the best method to incorporate the new color scheme in order to observe if the intensity channel would provide better contrast then the red channel that Dr. Aly originally used. Also, I needed to determine how to extend the algorithm to other datasets and I did so by proportionally altering the camera parameters based on the image resolution. In order to achieve the road sign and vehicle detection feature, I researched the Computer Vision technique, blob detection, and implemented it into the algorithm.