20 February 2022 in Biocybernetics and Biomedical Engineering
1 January 2021 in IEEE Access
Image data contain spatial information only, thus making two-dimensional (2D) Convolutional Neural Networks (CNN) ideal for solving image classification problems. On the other hand, video data contain both spatial and temporal information that must be simultaneously analyzed to solve action recognition problems. 3D CNNs are successfully used for these tasks, but they suffer from their extensive inherent parameter set. Increasing the network’s depth, as is common among 2D CNNs, and hence increasing the number of trainable parameters does not provide a good trade-off between accuracy and complexity of the 3D CNN. In this work, we propose Pooling Block (PB) as an enhanced pooling operation for optimizing action recognition by 3D CNNs. PB comprises three kernels of different sizes. The three kernels simultaneously sub-sample feature maps, and the outputs are concatenated into a single output vector. We compare our approach with three benchmark 3D CNNs (C3D, I3D, and Asymmetric 3D CNN) and three datasets (HMDB51, UCF101, and Kinetics 400). Our PB method yields significant improvement in 3D CNN performance with a comparatively small increase in the number of trainable parameters. We further investigate (1) the effect of video frame dimension and (2) the effect of the number of video frames on the performance of 3D CNNs using C3D as the benchmark.
18 November 2024 in IEEE Access
Text-based person re-identification (re-ID) is an emerging research domain in multi-camera surveillance systems. It involves identifying individuals across non-overlapping camera views using textual descriptions as queries. Although text-based person re-ID holds significant potential for surveillance systems, its practical applications are limited by the high computational demands of existing algorithms. This is because most state-of-the-art algorithms prioritize identification accuracy over resource efficiency. Therefore, surveillance systems based on these algorithms rely heavily on centralized architectures, where a central application server aggregates videos from camera nodes, generates galleries from the videos, and performs person identification. However, such centralized systems face challenges in large-scale deployments, including scalability issues, high bandwidth utilization, and processing bottlenecks. This paper presents a decentralized approach to text-based person re-ID to overcome these limitations. First, it introduces U-TextReIDNet, a resource-efficient model designed to identify persons in multi-person images on the NVIDIA Jetson Nano embedded board. U-TextReIDNet achieves Top-1 accuracies of 54.02% on the CUHK-PEDES dataset and 38.45% on the RSTPReid dataset. With only 38.77 million parameters, U-TextReIDNet is significantly smaller than most existing text-based person re-ID models. Using U-TextReIDNet, we implement a decentralized system that distributes the person identification task across camera nodes, transmitting only videos containing persons of interest to the command station. Additionally, we developed a prototype of this decentralized approach and conducted performance and usability tests using real human subjects. The prototype successfully performs real-time person re-ID, reduces bandwidth usage, improves system scalability, and eliminates processing bottlenecks.