Supervised and Unsupervised Learning Approaches for Tracking Moving Vehicles
R. Venkatesan, A. Balaji Ganesh · 2014
The paper presents both supervised and unsupervised learning approaches for the real time detection and tracking of moving vehicles. Learned template matching is based on the extraction of features such as position, color, shape and size. Video surveillance system consists of template generation and color pattern matching which is implemented as well as tested using both MATLAB and NI LabVIEW. It is found that visual saliency based unsupervised learning approach could be easily implemented and successfully integrated with camera modules in NI LabVIEW than MATLAB. Real-time performance validation of both environments is done by calculating learning time and elapsed time. The experimental results show that the unsupervised saliency algorithm is found to be robust to noise and blur or changes occurred in brightness or contrast levels. Visual attention model (saliency algorithm) in content-based image retrieval are employed to retrieve the interest points of video frame with the idea that these regions hold the best bottom-up features such as color, intensity, motion detection and orientation. These features are then typically utilized for future retrieval/classification tasks rather than using entire captured frame thus minimizing computational resources.