원문정보
초록
영어
One new pedestrian detection method integrating static high-level features and movement features based on convolution nerve network is proposed in this paper. During the phase of unsupervised deep learning of pedestrian features, the hierarchical static features of pedestrians are extracted from the low to the high with convolution nerve network; the pedestrian movement features are obtained through mean value approach of rectangular block pixel difference. During the logic regression recognition phase, static features and movement features are integrated. The results show that the pedestrian detection algorithm of convolution nerve network integrating movement features greatly improve the pedestrian detection performance under complicated background.
목차
1. Introduction
2. Deep Learning of Pedestrian Characteristics
2.1. Convolution Nerve Network
2.2. Non-Linear Transformation and Pooling
2.3. Movement Feature Extraction
2.4. Supervised Training in Combination with Movement Features
3. Experiment Result and Analysis
3.1. Experiment Data and Environment
3.2. Evaluation Standard
3.3. Contrast of Detection Precision
4. Conclusions
Reference