Successive Human Tracking and Posture Estimation with Multiple Omnidirectional Cameras

Akama, Shunsuke; Matsufuji, Akihiro; Sato-Shimokawara, Eri; Yamamoto, Shoji; Yamaguchi, Toru

doi:10.1109/taai.2018.00019

Cited by 3 publications

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They tested the algorithm on a real-time video surveillance system and claimed very low false alarms and missing detection. Akama et al [15] have used multiple omnidirectional cameras and a machine-learning algorithm to develop a method for human tracking and posture evaluation. By using two omnidirectional cameras, the trajectory of Head and foot position could be obtained continually to identify the behavior of residents.…”

Section: Literature Reviewmentioning

confidence: 99%

A Deep Learning based Human Detection and Tracking for Security Surveillance Systems

Irshad

Asif

Hassan

et al. 2023

ACE

View full text Add to dashboard Cite

All around the world, the crime rate has been increasing day by day, causing a rise in security issues. Closed-Circuit Television (CCTV) cameras have been installed throughout the world with the aim of decreasing crime and increasing public safety. The usage of CCTV cameras helps to increase crime detection accuracy significantly. Daily, a considerable amount of data has been recorded through CCTV cameras. Detection and recognition of culprits in the recorded data is a challenging task as it takes a lot of time, and human interaction is also involved. So, there is a need to develop a system that performs real-time detection and tracking of humans. This paper proposes a human detection and tracking system based on deep learning that assigns a unique ID to humans who enter the video scene. Multi-Task Cascaded Convolutional Neural Networks (MTCNN) and FaceNet models are used to achieve the desired target. The MTCNN model is trained on the WIDER SPACE dataset to perform human detection. FaceNet is used for human identification that is trained on the LFW dataset. The proposed system has been evaluated on 50 video sequences captured in different environments and achieved 97% average accuracy.

show abstract

Section: Literature Reviewmentioning

confidence: 99%