MODN - CHALLENGES AND EMERGING SOLUTIONS FOR MULTI OBJECT DETECTION ACROSS MULTIPLE NON-OVERLAPPING CAMERAS IN REAL TIME ENVIRONMENTS

Abstract

A crucial and developing research area, multi-object identification in multi-camera networks with non-overlapping fields of view serves as the foundation for applications including traffic analysis, intelligent surveillance, and smart city infrastructure. Non-overlapping camera networks present unique difficulties since there is no spatial or temporal continuity as objects move between camera perspectives, in contrast to conventional multi-camera systems with overlapping coverage. Due to the substantial changes in object appearance caused by differences in camera angles, lighting conditions, occlusions, and ambient factors, this discontinuity makes important tasks like object re-identification, data association, and trajectory reconstruction much more difficult. Additionally, occlusion events that happen within individual cameras as well as throughout the network make identity tracking even more problematic. Sophisticated feature extraction techniques and reliable re-identification systems that can deal with appearance changes, perspective shifts, and partial occlusions are required to handle these complications. This paper provides a thorough examination of current approaches, issues, and new developments in the field of cross-camera object tracking and identification. The effectiveness of conventional methods including data association techniques, graph-based optimization, and Kalman filtering in improving tracking consistency across non-overlapping views is investigated. Concurrently, considerable gains in feature representation and cross-view object matching have been shown by developments in deep learning, including convolutional neural networks and transformer-based architectures. Network calibration is an essential precondition for successful multi-camera tracking, and it presents particular difficulties in non-overlapping setups. A variety of calibration methods are thoroughly investigated, including automated calibration pipelines like CamMap, large-scale target placement, mirror-based systems, and the use of natural landmarks. Furthermore, computationally efficient solutions are required for real-time processing restrictions in large-scale deployments, which has led to the adoption of edge computing methodologies and distributed processing paradigms. Another fundamental requirement is precise synchronization across cameras, which is essential for ensuring accurate temporal alignment of multi-camera data streams. This paper reviews methodologies designed to mitigate synchronization drift and maintain consistent timestamps across heterogeneous camera networks. With the increasing adoption of smart city technologies and autonomous systems, there is a growing demand for scalable, robust multi-camera solutions capable of seamless object detection, tracking, and identity maintenance across environments with sparse camera placement. This study, titled challenges and emerging solutions for multi-object detection across multiple non-overlapping cameras in real-time environments, consolidates research developments, identifies existing limitations, and outlines prospective research directions. Key future advancements include integrating multi-modal data for improved recognition, leveraging self-supervised learning for adaptive feature extraction, and incorporating contextual environmental information to enhance object re-identification in non-overlapping camera networks. By addressing these challenges, this research contributes towards the realization of real-time, high-accuracy, and scalable cross-camera multi-object detection systems, paving the way for next-generation surveillance and intelligent monitoring solutions.