Estimating traffic condition using just a single image
Yao Bin Then, Yong Haur Tay, Wing Teng Ho · 2013
Accurate and fast information acquisition on traffic condition is vital to the urban drivers and city management. Today, most of the computer vision-based techniques in traffic condition monitoring perform on video stream, in which requires high networking bandwidth to transfer the video stream to the processing unit. In this paper, we present a simple yet effective adaptive technique that is able to estimate the traffic condition by just using a single image. The system is based on FAST corner detection and SURF keypoint descriptor, and multilayer perceptron (MLP). A video input is needed only during the training phase; however, no manual annotation is needed to provide teaching signals to the multilayer perceptron. Once the MLP is trained, the system is able to estimate the traffic condition by using one single image. We evaluate the system on a few real-world datasets under different illumination and traffic conditions, and obtain very positive results.