Person Search by Queried Description in Vietnamese Natural Language

Abstract

Nowadays, surveillance camera systems are widely deployed today from public places to private houses. This leads to huge image databases. Recent years have witnessed a significant improvement of surveillance video analysis, especially for person detection and tracking. However, finding the interested person in these databases is still very challenging issue. A majority of the existing person search methods bases on the assumption that the example image of the person of interest is available. This assumption is however not always satisfied in practical situations. Therefore, recently, person search by using query of natural language description has attracted the attention of researchers. However, they mainly dedicate to queried descriptions in English. In this paper, we propose a person search method with query of Vietnamese natural language. For this, Gated Neural Attention - Recurrent Neural Network (GNA-RNN) is employed to learn the affinity from pairs of description and image and then to estimate the similarity between query and images in the database. To evaluate the effectiveness of the proposed method, extensive experiments have performed in two datasets: CUHK-PEDES with description translated in Vietnamese and our own collected dataset named VnPersonSearch. The promising experimental results show the great potential of the proposed method.