Robust Real Time Moving People Detection in Surveillance Scenarios

Abstract

Esta es la versión de autor de la comunicación de congreso publicada en: This is an author produced version of a paper published in:

People detection based on appearance and motion models

“…[6] uses a person model based on HOG (histograms of oriented gradients) descriptors and an SVM classifier, [11] makes use of shape representation with the generative ISM framework, [19] uses an ellipse model and a silhouette fitting algorithm and [20] performs the classification by similarity with silhouettes stored in a codebook. On the other hand, there are methods based on combination of multiple parts [3,9,18,8]. [3] trains multiple detectors for anatomically defined body parts which are then combined using pictorial structures, [9] performs an analysis of concavity and convexity of the silhouette to identify different body parts, [18] tries to identify the characteristic edges of a human body and to generate four edge models (body, head, torso and legs), each model is trained using a nested Adaboost cascade structure and [8] proposes a realtime adaptation of the work presented in [18].…”

“…On the other hand, there are methods based on combination of multiple parts [3,9,18,8]. [3] trains multiple detectors for anatomically defined body parts which are then combined using pictorial structures, [9] performs an analysis of concavity and convexity of the silhouette to identify different body parts, [18] tries to identify the characteristic edges of a human body and to generate four edge models (body, head, torso and legs), each model is trained using a nested Adaboost cascade structure and [8] proposes a realtime adaptation of the work presented in [18].…”

Post-processing approaches for improving people detection performance

“…There are also some approaches that try to combine both approaches together [43], [44]. In any case, the result of this stage is the location and dimension (bounding box or blob) of the different objects candidates to be a person.…”

A Real-Time Human Detection System for Video

“…Breitenstein et al [19] utilize an off-line trained pedestrian detector [1] and online trained, instance-specific classifier via online boosting [20] for multi-person trackingby-detection. Another frequently used technique in video is background modeling, and most real-time systems [21], [22] depend on it for a fast speed. For QVGA videos, the systems [21], [22] have achieved real-time performance.…”

“…Another frequently used technique in video is background modeling, and most real-time systems [21], [22] depend on it for a fast speed. For QVGA videos, the systems [21], [22] have achieved real-time performance. But for VGA videos, the background modeling itself costs much time, making real-time processing challenging.…”