[Paper Review] People Counting in Crowded and Outdoor Scenes using a Hybrid Multi-Camera Approach

Fabio Dittrich, Luiz S. Oliveira+2 more|arXiv (Cornell University)|Apr 2, 2017

Video Surveillance and Tracking Methods31 references17 citations

TL;DR

This paper proposes a hybrid multi-camera approach for people counting in crowded outdoor scenes, combining direct detection using head detectors (SVM and Adaboost) and an indirect method based on corner point analysis and statistical modeling. The indirect method achieves a 46.7% improvement over baseline methods on the PETS2009 benchmark, outperforming the direct method due to reduced sensitivity to occlusion and segmentation errors.

ABSTRACT

This paper presents two novel approaches for people counting in crowded and open environments that combine the information gathered by multiple views. Multiple camera are used to expand the field of view as well as to mitigate the problem of occlusion that commonly affects the performance of counting methods using single cameras. The first approach is regarded as a direct approach and it attempts to segment and count each individual in the crowd. For such an aim, two head detectors trained with head images are employed: one based on support vector machines and another based on Adaboost perceptron. The second approach, regarded as an indirect approach employs learning algorithms and statistical analysis on the whole crowd to achieve counting. For such an aim, corner points are extracted from groups of people in a foreground image and computed by a learning algorithm which estimates the number of people in the scene. Both approaches count the number of people on the scene and not only on a given image or video frame of the scene. The experimental results obtained on the benchmark PETS2009 video dataset show that proposed indirect method surpasses other methods with improvements of up to 46.7% and provides accurate counting results for the crowded scenes. On the other hand, the direct method shows high error rates due to the fact that the latter has much more complex problems to solve, such as segmentation of heads.

Motivation & Objective

• Address the challenge of accurate people counting in dense, outdoor, and occlusion-prone environments where single-camera systems fail.
• Overcome limitations of single-view counting by fusing information from multiple camera perspectives to reduce occlusion and improve scene coverage.
• Develop and compare two distinct approaches: a direct method relying on individual head detection and an indirect method using group-level features and statistical modeling.
• Improve counting accuracy by incorporating a distance-weighting scheme to compensate for perspective distortion in multi-view setups.
• Evaluate the proposed methods on a real-world benchmark (PETS2009) to demonstrate robustness and scalability in complex surveillance scenarios.

Proposed method

• Use two calibrated cameras to capture overlapping views of a scene, enabling multi-view fusion to reduce occlusion and improve visibility.
• Implement a direct counting method using two head detectors: one based on Support Vector Machines (SVM) and another on Adaboost Perceptron, trained on head and non-head images.
• Apply a cascaded classifier architecture in the Adaboost Perceptron to improve detection efficiency and reduce false positives.
• For the indirect method, extract corner points from foreground groups of people and use a least-squares linear model to estimate group size based on area, perimeter, and edge features.
• Apply a perspective compensation weighting scheme to correct for distance-related distortion in the scene, improving accuracy across varying camera-to-person distances.
• Fuse results from both views using homography-based correspondence to estimate the total number of people in the entire scene, not just individual frames or views.

Experimental results

Research questions

• RQ1 Can multi-camera fusion significantly improve people counting accuracy in crowded outdoor scenes compared to single-camera systems?
• RQ2 How does the performance of a direct detection-based method compare to an indirect statistical modeling approach in high-density, occlusion-prone environments?
• RQ3 To what extent does incorporating a distance-weighting scheme improve the accuracy of multi-view people counting?
• RQ4 What are the key challenges in establishing reliable point correspondence between multiple camera views for scene-level counting?
• RQ5 How do camera placement and height affect the reliability and accuracy of multi-view people counting systems?

Key findings

• The indirect method achieved a 46.7% improvement over baseline methods on the PETS2009 benchmark, demonstrating superior robustness in crowded scenes.
• The indirect method outperformed the direct method significantly, with the direct method suffering from high error rates due to segmentation and detection failures in dense crowds.
• The Adaboost Perceptron detector showed poor performance on both views, with average errors exceeding 20 per frame, indicating limitations in training data and false alarm control.
• The indirect method’s use of a perspective-weighting scheme enhanced accuracy by compensating for distance-related distortions in the scene.
• Camera height and positioning are critical: optimal placement (5–7 meters high) and non-overlapping field-of-view coverage (e.g., avoiding cameras that follow the walkway) significantly improved results.
• Temporal consistency, such as detecting heads persisting across multiple frames, could improve detection reliability and reduce false positives in future work.