基于注意力评分函数的行人重识别研究

Pedestrian reidentification study based on the attention scoring function

行人重识别也称跨境追踪,旨在弥补固定摄像头的视觉局限。针对行人图像容易出现遮挡、视觉与姿态的不同变化、光照变化等影响而出现难以区分行人的状况,最终导致行人重识别准确率低的问题,文章提出了一种使用点积得到计算效率更高的注意力评分函数进行检测的方法。实验结果表明,在注意力机制的加持下,该模型能够有效地增强行人图像特征等关键词的提取,进一步提高模型的鲁棒性,有效地满足实际需要。

Person re-identification,also known as cross-border tracking,aims to make up for the visual limitations of fixed cameras,and this paper proposes a method for detecting pedestrians by using the attention scoring function with higher computational efficiency by using the dot product to obtain a more efficient attention scoring function for detection.The experimental results show that with the support of the attention mechanism,the model can effectively enhance the extraction of keywords such as pedestrian image features,further improve the robustness of the model,and effectively meet the practical needs.