Abstract
Intelligent Transportation Systems (ITS) allow us to have high quality traffic information to reduce the risk of potentially critical situations. Conventional image-based traffic detection methods have difficulties acquiring good images due to perspective and background noise, poor lighting and weather conditions. In this paper, we propose a new method to accurately segment and track vehicles. After removing perspective using Modified Inverse Perspective Mapping (MIPM), Hough transform is applied to extract road lines and lanes. Then, Gaussian Mixture Models (GMM) are used to segment moving objects and to tackle car shadow effects, we apply a chromacity-based strategy. Finally, performance is evaluated through three different video benchmarks: own recorded videos in Madrid and Tehran (with different weather conditions at urban and interurban areas); and two well-known public datasets (KITTI and DETRAC). Our results indicate that the proposed algorithms are robust, and more accurate compared to others, especially when facing occlusions, lighting variations and weather conditions.
Highlights
Accurate and real-time traffic information detection in different weather conditions has become a significant problem
As shown in the “Tables 2 and 3” and “Figs 3, 8, 9, 10, 13 and 14”, Modified Inverse Perspective Mapping (MIPM) performance is comparable with Inverse Perspective Mapping (IPM) and Homography method in the first frames, but it is different in the following ones
We have proposed a robust method for extracting real information from traffic cameras
Summary
Accurate and real-time traffic information detection in different weather conditions has become a significant problem. We propose a vision based, real-time traffic information detection algorithm that uses modified inverse perspective mapping MIPM This method is recommended to remove the perspective from images to accurately detect vehicles in various weather conditions. Our simulation results verified the better performance of the proposed method compared to similar works in delectability and traceability under different weather conditions, perspective and background noise, shadows and lighting transitions, all of which are difficulties conventional traffic detection methods have to deal with. This paper is organized as follows: section provides the details to extract real traffic information through different techniques like Modified Inverse Perspective Mapping, Hough transform and Gaussian Mixture Models to detect the vehicle. Using Gaussian Mixture Model and Chromacity-based Method we segmented the cars, removed the shadows, and the vehicles areas were calculated This was used as a measure to evaluate the performance of IPM, homography and MIPM. It can be inferred that MIPM outperforms IPM and Homography “Table 2” as when we use MIPM:
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
