Bus Detection for Adaptive Traffic Signal Control

DISCLAIMER The contents of this report reflect the views of the authors, who are responsible for the facts and the accuracy of the information presented herein. This document is disseminated under the sponsorship of the U.S. 1 Qualitative results. (a) A tiny bus detected with score 1.000. (b) An occluded bus detected with score 1.000 shown in green and a false positive with score 0.926 shown in red. Figure 7 suggests that this is a very low false positive score.. 8 2 The dataset is challenging because of noise, large change in illumination , heavy occlusions and large variation in object size in image. The yellow line indicates the boundary (based on the upper y coordinate) between buses near the intersection and buses far from the intersection. 9 3 Precision Recall for Laboratory Experiments. " ESVM+Domain " is Exemplar SVM operating on a cropped image, with two zones, without image flip add detection, without model sampling, without a HOG pyramid, with weibull calibration. " No Zones " is the same except that the image is not divided into two zones. " No Zones and No Lane Cropping " is the same as " No Zones " except that the image is not cropped to focus on the lane. " No Calibration " is the same as " ESVM+Domain " but does not use any form of exemplar calibration, while " Platt Calibra