Search for: All records

After a disaster strikes an urban area, damage to the façades of a building may produce dangerous falling hazards that jeopardize pedestrians and vehicles. Thus, building façades must be rapidly inspected to prevent potential loss of life and property damage. Harnessing the capacity to use new vision sensors and associated sensing platforms, such as unmanned aerial vehicles (UAVs) would expedite this process and alleviate spatial and temporal limitations typically associated with human-based inspection in high-rise buildings. In this paper, we have developed an approach to perform rapid and accurate visual inspection of building façades using images collected from UAVs. An orthophoto corresponding to any reasonably flat region on the building (e.g., a façade or building side) is automatically constructed using a structure-from-motion (SfM) technique, followed by image stitching and blending. Based on the geometric relationship between the collected images and the constructed orthophoto, high-resolution region-of-interest are automatically extracted from the collected images, enabling efficient visual inspection. We successfully demonstrate the capabilities of the technique using an abandoned building of which a façade has damaged building components (e.g., window panes or external drainage pipes). 

Abstract Health monitoring of civil infrastructures is a key application of Internet of things (IoT), while edge computing is an important component of IoT. In this context, swarms of autonomous inspection robots, which can replace current manual inspections, are examples of edge devices. Incorporation of pretrained deep learning algorithms into these robots for autonomous damage detection is a challenging problem since these devices are typically limited in computing and memory resources. This study introduces a solution based on network pruning using Taylor expansion to utilize pretrained deep convolutional neural networks for efficient edge computing and incorporation into inspection robots. Results from comprehensive experiments on two pretrained networks (i.e., VGG16 and ResNet18) and two types of prevalent surface defects (i.e., crack and corrosion) are presented and discussed in detail with respect to performance, memory demands, and the inference time for damage detection. It is shown that the proposed approach significantly enhances resource efficiency without decreasing damage detection performance.