Abstract
Despite being widely adopted and rigorously followed in many successful pedestrian detectors, the original HOG (Histogram of Oriented Gradients) descriptor are in fact NOT optimally tuned for pedestrian detection. To address this issue, we quantitatively investigate the interplay among different HOG parameters, in particular that among the cell size, aspect ratio, and detection window size, which makes it possible to jointly tune these parameters to achieve better pedestrian detection performance. In addition, we extend the training procedure of the original HOG-based detector of Dalal et al. through presenting an automatic positive sample generation algorithm, introducing LSVM (Latent SVM) to iteratively optimize the positive training samples, and adopting a hard negative mining method. To verify the effectiveness of our improved detector, we conduct extensive experiments on INRIA Person, TUD-Brussels and Caltech Pedestrians datasets. On all these datasets, our detector outperforms significantly the original HOG detector of Dalal et al.
Sign-in/Register to access full text options 
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 callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
