Pedestrian Attribute Recognition in Elevator

Pengyuan Zhang,Yi Shi

DOI: https://doi.org/10.1109/icaica54878.2022.9844591

2022-01-01

Abstract:Recognizing pedestrian attributes in elevator video surveillance plays a critical role in community safety and business decision. Existing methods have paid little attention to such scenarios, in which the essential facial information usually cannot be obtained and used for recognizing pedestrian attributes. Considering that low-level attributes have strong correlations with high-level attributes, e.g., gender can be predicted with the combination of hair length and dressing style, in this paper, we constructed a large multi-label dataset for pedestrian attribute recognition in elevator and proposed an end-to-end multi-label model to improve attribute recognition. This model consists of 1) channel attention mechanism, which was used to ignore the irrelevant background information, 2) local feature blocks extraction which can fully extract the local feature of the image, 3) multiple classifiers which can accelerate the speed of reasoning by avoiding repeated computation of features in the shared layer. In the inference stage, we used FPN and feature map dimensionality reduction to predict attributes. Compared with the baseline model, the performances on our dataset and the other two large-scale pedestrian attribute datasets, i.e., PETA and MarKet1501, demonstrate the effectiveness of the proposed method.