TensorView: A New Method for Data Collection Using Google Street View and Image Classification Technologies
Satellite and aerial imagery, though constantly improving, is typically of insufficient resolution and at the wrong angle to detect features of the streetscape. This often results in information that is of interest not being available, necessitating a costly and time-consuming manual survey. This leaves open space for a "third category" of data-gathering methodologies that utilize third-party data and recent advances in technology. In this paper, we introduce TensorView, a novel tool that leverages Google Street View imagery and cutting-edge image classification technology in order to decrease surveying workload on GIS practitioners. As a proof of concept, and to develop the workflow, we apply this tool to a task that is poorly suited to traditional remote-sensing techniques - the detection of abandoned and blighted buildings in an urban area. This example study reveals that, even in proof-of-concept form, the TensorView tool shows great promise in the pre-survey role, demonstrating an overall error rate of only 11% on an extremely challenging task and requiring significantly less human labour to accomplish it.