RESEARCH ON THE USE OF POSTGRESQL IN A HUMAN RE-IDENTIFICATION INFORMATION SYSTEM

This paper investigates the application of the PostgreSQL database management system in the context of human re-identification systems based on computer vision technologies. Considering the growing need for efficient storage and processing of large-scale vector data obtained through deep neural networks, the issue of optimizing databases for fast and accurate matching of individuals' vector features arises. The study examines the specifics of storing high-dimensional vector representations in PostgreSQL using the pgvector extension, which provides efficient similarity search capabilities using metrics like cosine similarity and Euclidean distance. A detailed experimental analysis of PostgreSQL's performance was conducted during the execution of nearest neighbor search operations (k-NN search) based on vector features obtained from the open dataset Market-1501. The speed and accuracy of PostgreSQL were compared with other popular DBMSs such as MySQL and MongoDB, utilizing similar data storage and indexing methods. The research results demonstrate that PostgreSQL with the pgvector extension provides significantly better performance, achieving an average query execution time of 35 ms and a throughput of 285 queries per second, surpassing the metrics of MySQL and MongoDB. Moreover, the use of PostgreSQL contributes to increased accuracy in re-identification, as confirmed by higher mAP (mean Average Precision) values (78% compared to 72% in MySQL and 70% in MongoDB). Factors influencing system performance are discussed, including indexing efficiency, query optimization, and system resource utilization. The paper also explores the scalability of PostgreSQL for processing even larger volumes of data and its integration with other tools to enhance performance. The obtained results confirm the feasibility of using PostgreSQL in human re-identification systems, which can have practical significance for developing high-performance and scalable security and monitoring systems. Future research may focus on optimizing search algorithms and integrating advanced machine learning technologies to further improve system accuracy and speed