What is a Hole? Discovering Access Holes in Disaster Rubble with Functional and Photometric Attributes

Abstract

The collapse of buildings and other structures in heavily populated areas often results in human victims becoming trapped within the resulting rubble. This rubble is often unstable, difficult to traverse, and dangerous for emergency first responders tasked with finding, stabilizing, and extricating entombed or hidden victims through access holes in the rubble. Recent work in scene mapping and reconstruction using photometric color and metric depth (RGB-D) data collected by unmanned aerial vehicles (UAVs) suggests the possibility of automatically identifying potential access holes into the interior of rubble. This capability would greatly improve search operations by directing the limited human search capacity to areas where access holes might exist. This paper presents a novel approach to automatically identifying access holes in rubble. The investigation begins by defining an access hole in terms that allow for their algorithmic identification as a potential means of accessing the interior of rubble. This definition captures the functional and photometric attributes of holes. From this definition, a set of hole-related features for detection is presented. Experiments were conducted using RGB-D data collected over a real-world disaster training facility using a UAV. Empirical evaluation suggests the efficacy of the proposed approach for successfully identifying potential access holes in disaster rubble. C 2015Wiley Periodicals, Inc.

Classifying Negative Objects With Neural Networks

“…The term "negative obstacle" has a widely accepted definition as areas where there is an extension/depression into a surface, such as a hole or cliff [6], [17], [18], [19], [20], [21], [22], [23], [24], [1], [25], [26], [27], [28], [29].…”

“…Kong et al 2015, [25], analyzes individual stereo image pairs with data types RGB and depth to determine if the full colour depth image contains a hole with dimensions large enough to support a human, this is known as an access hole. Refining the implementation of the work proposed in [32], [25] creates a new dataset using an Asus Xtion colour-depth camera mounted on a UAV instead of the data set collected by [15] using a UAV-mounted Microsoft Kinect colour-depth camera, in an attempt to account for the shortcomings if the colour-depth camera mentioned in both [15] and [32]. Basing their definition on geometry and appearance taken in stereo, they are able to calculate the size of the hole and determine if it is large enough to support a human being.…”

“…Each of the thresholds that a potential access hole must pass were defined by Kong et al based on the appearances of access holes in the dataset and thereby forces the definition to be exclusive to a fault. [25] poses the assumption that all access holes will be poorly illuminated in comparison to its surroundings based on research done in [32]. Since the data was collected with colour-depth camera (intended for indoor use only), sunlight was an issue that skewed the data set, which is the same problem found in [15] (demonstrated in Figure 2.1).…”

“…[33] improves and extends upon [25] which helps reduce the problem of having poor depth data. [33] modifies how much feature scores affect the final total based on how pronounced the feature is calculated to be, and if the algorithm produces a positive result, images taken from the same area are searched to see if they also produce a similar result.…”

“…[33] modifies how much feature scores affect the final total based on how pronounced the feature is calculated to be, and if the algorithm produces a positive result, images taken from the same area are searched to see if they also produce a similar result. While this method shows increased average precision in detection by 22% compared [25] and was able to reach 100% hole recall at the cost of speed, they also exploit brightness change as key attributes of an access hole. With the construction of a RP, shadowing over areas that do not contain an access hole could result in a false positive.…”