PROJECT TITLE :

Discriminative Weighted Sparse Partial Least Squares for Human Detection

ABSTRACT:

Channel feature detectors have shown nice advantages in human detection. But, a massive pool of channel features extracted for human detection typically contains several redundant and irrelevant features. To deal with this issue, we propose a robust discriminative weighted sparse partial least square approach for feature selection and apply it to human detection. Unlike partial least squares (PLS), which is a straightforward dimensionality reduction technique, we have a tendency to propose using sparse PLS to achieve feature choice. Furthermore, in order to get a strong latent matrix, we formulate a discriminative regularized weighted least sq. problem, where a discriminative term is incorporated to effectively distinguish positive samples from negative samples. A robust sparse weight matrix is trained based mostly on the latent matrix and used for feature choice. Finally, we use the chosen channel options to train the boosted decision trees and incorporate the weights of selected features with every tree. The human detector trained by the chosen options will preserve high robustness and discriminativeness. Experimental results on some difficult human knowledge sets demonstrate that the proposed approach is effective and achieves state-of-the-art performance.