Lane Change Detection using Fusion of IPM and SVM

U. Qayyum and M.Y. Javed (Pakistan)


Inverse Perspective Mapping, Support Vector Machine, Lane marker detection and Lane Localization.


Lane marker detection and lane departure warning systems are well-researched areas of computer vision. This research paper deals with the implementation of IPM and SVM techniques for lane change detection and localization in order to provide timely assistance to the vehicle drivers. During the preprocessing phase, the algorithm of the developed system utilizes IPM for geometric transformation to remove the perspective effect from the acquired scene. The remapped image of IPM represents the lane markers much brighter than their surroundings for better detection and linking. Lane localization and transition detection has been achieved using SVM method by establishing hyper-plane through the kernel functions. The developed system was trained on 12 images of different road scenarios, four images for every lane model of Lahore-Islamabad (Pakistan) motorway. Experimental results of the algorithm has demonstrated that the developed system provides lane change detections accuracy up to 94%, which is much higher compared to individual implementation of IPM (i.e. 64%), SVM (i.e. 82%) and HMM (i.e. 70%). This algorithm is capable of processing 12 frames per second, which makes it perfect for real time applications and it can be utilized more effectively in the intelligent navigational assistance and accident avoidance systems.

Important Links:

Go Back