Automated Structural Health Monitoring of Pavements, Using Google Street View Images

The Department of Transportation traditionally employed a structural health monitoring approach that involved training personnel to report the existing conditions of roadways. However, this approach is accompanied by huge costs that range from onboarding, vehicular equipment & maintenance cost, environmental costs from CO₂ emission, safety costs, and the cost of errors made by personnel as each inspector would exercise his/ her discretion in analyzing the condition of the roadway. Therefore, it is necessary to adopt a more efficient and effective approach to SHM. The incorporation of google street view images [‘GSV’ a Google roadway street data repository, with high-resolution images], can minimize the safety risk, the operational costs, and provide an accurate assessment of pavement conditions. 

Automated SHM involves the use of artificial intelligence [AI]. Over the past decade, researchers have analyzed the uses, importance, and benefits of engineering to efficiently compute complex mathematical/ engineering challenges. Artificial intelligence can be categorized into Symbolic AI [an aspect that uses logic like searching and solving problems using rule base programs], and Machine learning [an aspect that deduces inferences from raw data using relational reasoning rather than from explicit statements]. This relation is drawn from neural networks with many feed-forward layers on large datasets convolutionally. Convolution in its simplest form is a neuron-like processing unit comprised of an input layer, hidden layers, and an output layer, which performs complex computations by utilizing fully connected layers with independent weights.

CNN is effective in feature extraction [5] and is widely used in computer vision tasks such as image classification, and object recognition. Therefore, our goal is to design and build a machine-learning model using GSV, to efficiently and effectively detect the presence of cracks on the roadway, if any. In this research, a novel technique utilizing GSV, as its primary image dataset repository for up-to-date roadway pavement conditions, and a novel architecture uniquely built to extract cracks is trained with GSV CubeMap processed images. Finally, our novel machine learning model will be compared to existing machine learning models, using industry metrics, to measure the robustness and effectiveness of our model prediction and accuracy.