Abstract
A video-based traffic monitoring system must be capable of working in various weather and illumination conditions. In this paper, we will propose an example-based algorithm for moving vehicle detection. Different from previous works, this algorithm learns from examples and does not rely on any a priori model for vehicles. First, a novel scheme for adaptive background estimation is introduced. Then, the image is divided into many small nonoverlapped blocks. The candidates of the vehicle part can be found from the blocks if there is some change in gray level between the current image and the background. A low-dimensional feature is produced by applying principal component analysis to two histograms of each candidate, and a classifier based on a support vector machine is designed to classify it as a part of a real vehicle or not. Finally, all classified results are combined, and a parallelogram is built to represent the shape of each vehicle. Experimental results show that our algorithm has a satisfying performance under varied conditions, which can robustly and effectively eliminate the influence of casting shadows, headlights, or bad illumination.
Original language | English |
---|---|
Pages (from-to) | 51-59 |
Number of pages | 9 |
Journal | IEEE Transactions on Vehicular Technology |
Volume | 56 |
Issue number | 1 |
DOIs | |
Publication status | Published - 1 Jan 2007 |
Keywords
- Principal component analysis (PCA)
- Statistical learning
- Support vector machine (SVM)
- Video-based traffic monitoring
ASJC Scopus subject areas
- Automotive Engineering
- Aerospace Engineering
- Applied Mathematics
- Electrical and Electronic Engineering