Using Computer Vision To Label And Search A Physical Space

Effective operation of a warehouse requires keeping track of the location of various assets within the physical environment. As various sensors are carried through the warehouse environment by operators, range data collected by the sensors over time can be used to reconstruct 2D and 3D representations of the space. This disclosure describes techniques to estimate the locations of Point-Of Interest (POIs) and Regions-Of-Interest (ROIs) within a physical environment such as a warehouse. The location estimates are generated using a combination of 2D visual search of images containing text labels and barcodes, 2D/3D environment reconstruction using sensor data, and estimated trajectory of sensors. Computer vision techniques are applied to visual data which is obtained from operational processes that generate images, such as feeds from stationary cameras, images from moving cameras, photos of the environment, etc.