A Survey on Visual Surveillance of Object Motion and Behaviors

Hu, W.; Tan, T.; Wang, L.; Maybank, S. · 2004 · OpenAlex-citations

DOI: 10.1109/tsmcc.2004.829274

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This 2004 survey paper by Hu et al. addresses the growing need for intelligent visual surveillance systems capable of automatically detecting, recognizing, and tracking objects in dynamic scenes. The research is motivated by the limitations of traditional passive video surveillance, which becomes ineffective as the number of cameras exceeds human monitoring capabilities. The authors focus specifically on applications involving humans and vehicles, such as access control, person-specific identification, crowd flux statistics, anomaly detection, and multi-camera interactive surveillance. The paper aims to provide a comprehensive review of the general framework of visual surveillance, organizing recent developments into a hierarchical structure to help researchers understand the state-of-the-art and identify future directions. The authors structure their review around a general processing framework consisting of environment modeling, motion detection, object classification, tracking, behavior understanding, human identification, and multi-camera data fusion. The paper details methods for motion detection, including environment modeling techniques like adaptive Gaussian estimation and background subtraction, as well as segmentation approaches such as temporal differencing and optical flow. Object classification is categorized into shape-based methods (using features like blob area and aspect ratio) and motion-based methods (exploiting periodicity in human gait). Tracking algorithms are divided into four types: region-based, active contour-based, feature-based, and model-based. The survey examines specific implementations, such as the W4 system for group tracking and the Pfinder system for 3D human modeling, while analyzing the merits and demerits of various algorithms regarding computational complexity, occlusion handling, and real-time performance. Key findings highlight the trade-offs between different surveillance techniques. Region-based tracking is effective for simple scenes but struggles with occlusion and clutter. Active contour-based methods offer better shape description and robustness to partial occlusion but are sensitive to initialization and computationally demanding. Feature-based tracking allows for real-time processing of multiple objects but may lack detailed pose information. Model-based tracking provides accurate 3D pose recovery but requires complex computational models. The paper also notes that while significant progress has been made in single-camera surveillance, challenges remain in handling occlusion, integrating 2D and 3D tracking, and fusing data from multiple sensors. The significance of this work lies in its systematic organization of visual surveillance technologies into low-, intermediate-, and high-level vision tasks. By categorizing methods and discussing their inter-component links, the paper provides a clear roadmap for newcomers and experts alike. It concludes by identifying critical future research directions, including improved occlusion handling, the combination of motion analysis with biometrics, anomaly detection and behavior prediction, content-based video retrieval, and remote surveillance. This survey serves as a foundational reference for understanding the evolution of computer vision applications in security and traffic management.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• distraction detection algorithms