[Paper Review] Moving Object Detection in Video Using Saliency Map and Subspace Learning

Yanwei Pang, Ye Li+2 more|arXiv (Cornell University)|Sep 30, 2015

Video Surveillance and Tracking Methods28 references3 citations

TL;DR

This paper proposes MODSM, a novel moving object detection method that integrates saliency maps into an incremental subspace learning framework to reduce false and missed alarms. By jointly optimizing for sparsity, low-rank structure, connectivity, and saliency, the method improves foreground detection in videos with challenging backgrounds, textureless regions, and non-meaningful motion, achieving state-of-the-art F1-score of 0.7711 on benchmark sequences.

ABSTRACT

Moving object detection is a key to intelligent video analysis. On the one hand, what moves is not only interesting objects but also noise and cluttered background. On the other hand, moving objects without rich texture are prone not to be detected. So there are undesirable false alarms and missed alarms in many algorithms of moving object detection. To reduce the false alarms and missed alarms, in this paper, we propose to incorporate a saliency map into an incremental subspace analysis framework where the saliency map makes estimated background has less chance than foreground (i.e., moving objects) to contain salient objects. The proposed objective function systematically takes account into the properties of sparsity, low-rank, connectivity, and saliency. An alternative minimization algorithm is proposed to seek the optimal solutions. Experimental results on the Perception Test Images Sequences demonstrate that the proposed method is effective in reducing false alarms and missed alarms.

Motivation & Objective

• To address high false alarm and missed alarm rates in traditional moving object detection due to noise, clutter, and textureless regions.
• To leverage saliency maps to guide foreground detection by emphasizing semantically meaningful moving objects over background variations.
• To unify sparsity, low-rank, connectivity, and saliency in a single optimization framework for robust video background modeling.
• To develop an incremental algorithm that outperforms batch methods in accuracy and adaptability to dynamic scenes.
• To reduce dependency on perfect saliency maps by making the method robust even when saliency maps are suboptimal.

Proposed method

• The method formulates a unified objective function combining saliency map, low-rank background matrix, sparse foreground, and connectivity constraints.
• It uses an alternating direction method of multipliers (ADMM)-based algorithm to solve the non-convex optimization problem iteratively.
• The saliency map is used as a spatial prior to suppress false positives in non-meaningful motion regions like waving leaves or ripples.
• Foreground is modeled as a sparse matrix, while background is represented as a low-rank matrix in a subspace learned incrementally over time.
• Connectivity constraints are applied to ensure spatial coherence of detected moving objects.
• The method is incremental, enabling real-time processing of video sequences without full re-computation.

Experimental results

Research questions

• RQ1 Can integrating a saliency map into subspace learning reduce false alarms from non-meaningful motion such as waving leaves or ripples?
• RQ2 Can the proposed method improve detection of textureless moving regions that are often missed by conventional methods?
• RQ3 How does the fusion of saliency with low-rank and sparse matrix decomposition affect detection performance compared to standalone subspace or saliency methods?
• RQ4 To what extent does the quality of the saliency map influence the final detection accuracy?
• RQ5 Can an incremental optimization framework achieve better performance than batch methods in moving object detection?

Key findings

• The proposed MODSM method achieved the highest average F1-score of 0.7711 across nine video sequences, outperforming RFDSA (0.7489) and DECOLOR.
• When the Lobby video—where the saliency map performance degraded—was excluded, MODSM’s F1-score improved to 0.7955, demonstrating sensitivity to saliency quality.
• On the Fountain video, MODSM achieved a true positive rate of 0.99 at a false positive rate of 0.05, outperforming RFDSA (0.935).
• Visual results show MODSM closely matches ground truth, while RFDSA and DECOLOR suffer from false positives (e.g., classifying escalators or leaves as foreground) and missed detections.
• The method effectively suppresses false alarms in scenes with dynamic backgrounds and performs well even with suboptimal saliency maps.
• The integration of saliency as a spatial prior significantly improves detection of moving objects with low texture, reducing missed alarms.