|Title:||Pedestrian detection in crowded scenes|
|Subject:||Hong Kong Polytechnic University -- Dissertations|
Pedestrian traffic flow
Pedestrians -- Mathematical models.
Image analysis -- Mathematical models
|Department:||Department of Electronic and Information Engineering|
|Pages:||[v], 51 leaves : ill. ; 31 cm.|
|Abstract:||The challenge in pedestrian detection lies on the wide variations in human appearance and poses that can occur in a sampled scene. Successful detection in crowded scenes with overlapping pedestrians and background clutter requires classifiers with sufficient complexity. This thesis will study an efficient feature for representing pedestrians, namely the histogram of oriented gradients (HOG), and will demonstrate a dense-scan multi-resolution pedestrian detector with improvements in terms of computation speed and accuracy. A multi-resolution cascade consisting of two layers of HOG detectors with different block sizes is developed, The first layer is a HOG classifier with a larger block size that can efficiently reject simple non-pedestrian samples. The second layer of the cascade consists of a HOG classifier with a smaller block size to increase detection accuracy. To further reduce the computation time while minimizing the memory constraints, nonnalized blocks are buffered to reduce the redundancy between the successive detection windows. Finally, a scheme will be proposed to reduce the computationally intensive square foot and inverse tan calculations required on the gradients in each pixel through using lookup tables. Experimental results show that the proposed method can reduce the computation time per frame by approximately 9.4%, and maintaining the same detection accuracy.|
Files in This Item:
|b23568124.pdf||For All Users (off-campus access for PolyU Staff & Students only)||7.65 MB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: