Аннотация
Evaluating pedestrians is critical when designing new systems for more sophisticated video monitoring and innovative computer vision-based security. Although convolutional neural networks successfully learn different features from images, modeling the pedestrians to the details needed for such tasks remains an issue. In this paper, we present a novel approach based on the multi-head deep learning model to identify the pedestrian attributes, which has merits in this method. Thus, the proposed approach yields better accuracy, robustness, and efficiency in comparison to state-of-the-art methods on a novel database consisting of pedestrian images. We also incorporate further pre-processing steps whereby data augmentation, transfer learning, and model ensembling are additional methods that are used to improve the generality and applicability of the approach proposed in this study. This framework empirically achieves promising accuracy and outperforms existing approaches, proving the usefulness and robustness of our idea in terms of pedestrian attribute recognition.