<title>Shape-based human detection for threat assessment</title>

Abstract

Detection of intrusions for early threat assessment requires the capability of distinguishing whether the intrusion is a human, an animal, or other objects. Most low-cost security systems use simple electronic motion detection sensors to monitor motion or the location of objects within the perimeter. Although cost effective, these systems suffer from high rates of false alarm, especially when monitoring open environments. Any moving objects including animals can falsely trigger the security system. Other security systems that utilize video equipment require human interpretation of the scene in order to make real-time threat assessment. Shape-based human detection technique has been developed for accurate early threat assessments for open and remote environment. Potential threats are isolated from the static background scene using differential motion analysis and contours of the intruding objects are extracted for shape analysis. Contour points are simplified by removing redundant points connecting short and straight line segments and preserving only those with shape significance. Contours are represented in tangent space for comparison with shapes stored in database. Power cepstrum technique has been developed to search for the best matched contour in database and to distinguish a human from other objects from different viewing angles and distances.

A Camera-Based Target Detection and Positioning UAV System for Search and Rescue (SAR) Purposes

“…In contrast, human detection based on color imagery has also been studied for many years. The research on developing a human detection method was conducted, which uses background subtraction, but pre-processing is required before a search mission [17]. Another method of human detection was presented that uses color images and models the human/flexible parts, then detects the parts separately [18].…”

Real Time Human Detection and Localization Using Consumer Grade Camera and Commercial UAV

“…One of them is by frame differencing where the algorithm performs calculations to check the difference in pixel intensity between consecutive frames. If the difference is substantial it can be said that there is something in the image [6]. The method of Mikolajczyk, Schmid, & Zisserman relies on identifying humans based on parts.…”

Bridging the semantic gap using Ranking SVM for image retrieval

“…Finally, the whole vertebra shape and the AO corner shape features are extracted and normalized using Min-Max normalization. The global shape features are geometric (elongation, eccentricity, roughness, and compactness), Fourier descriptors with complex coordinates [15], Fourier descriptors with Centroid Contour Distance Curve [15], Fourier Coefficient of Fourier Expansion of Bent function [16], and moment invariants. The local shape features are turn angle and Distance Across the Shape [17].…”