Large Scale Business Discovery from Street Level Imagery

We address the challenging problem of detecting business store fronts in street level imagery. Business store fronts are a challenging class of objects to detect due to high variability in visual appearance. Inherent ambiguiti es in visually delineating their physical extents, especiall y in urban areas, where multiple store fronts often abut each other, further increases complexity. We posit that traditional object detection approaches such as those based on exhaustive search or those based on selective search followed by post-classification are ill suited to address this problem due to these complexities. We propose the use of a Multibox [4] based approach that takes as input image pixels and directly outputs store front bounding boxes. Thi s end-to-end learnt approach instead preempts the need for hand modelling either the proposal generation phase or the post-processing phase, leveraging large labelled trainin g datasets. We demonstrate our approach outperforms the state of the art detection techniques with a large margin in terms of performance and run-time efficiency. In the evaluation, we show this approach achieves human accuracy in the low-recall settings. We also provide an end-to-end eval uation of business discovery in the real world.