Abstract
Better features have been driving the progress of pedestrian detection over the past years. However, as features become richer and higher dimensional, noise and redundancy in the feature sets become bigger problems. These problems slow down learning and can even reduce the performance of the learned model. Current solutions typically exploit dimension reduction techniques. In this paper, we propose a simple but effective feature selection framework for pedestrian detection. Moreover, we introduce occluded pedestrian samples into the training process and combine it with a new feature selection criterion, which enables improved performances for occlusion handling problems. Experimental results on the Caltech Pedestrian dataset demonstrate the efficiency of our method over the state-of-art methods, especially for the occluded pedestrians.
Highlights
Pedestrian detection plays an essential role in surveillance and security, and in computer vision for autonomous driving
We extended [26] in the following ways: (1) instead of focusing only at the occlusion handling problem, we design a new feature selection framework which improves the general detection performance, and (2) extra experiments with the Convolutional Neural Network (CNN) features and combined features are conducted to demonstrate the effectiveness of our method
We have proposed a robust feature selection framework for pedestrian detection
Summary
Pedestrian detection plays an essential role in surveillance and security, and in computer vision for autonomous driving. We obtain more optimal features by analyzing the models trained in the early training stages, without increasing the overall training time Another challenge in pedestrian detection is occlusion, which is currently handled in the following two ways: (1) Deformable Part Models (DPM) [21] and related methods; and (2) training a set of occlusion-specific models. We introduce the occluded pedestrian samples into training, and propose a new feature selection criterion, which obviously improves the occlusion handling ability of the model. We extended [26] in the following ways: (1) instead of focusing only at the occlusion handling problem, we design a new feature selection framework which improves the general detection performance, and (2) extra experiments with the CNN features and combined features are conducted to demonstrate the effectiveness of our method.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have