Autocalib: automatic traffic camera calibration at scale

Large scale camera installations are the becoming increasingly common in emerging smart cities. Though deployed primarily for surveillance, calibrating these cameras can allow them to measure real-world distances. This enables a broad spectrum of novel applications such as identifying speeding vehicles, city road planning , etc. Today, camera calibration is a tedious manual process and therefore not scalable to large camera installations. In this demo, we present AutoCalib, a system for scalable automatic calibration of traffic cameras. AutoCalib employs deep learning to identify selected key-point features from car images and uses a novel filtering and aggregation algorithm to automatically produce a robust estimate of the camera calibration parameters from just hundreds of samples. AutoCalib is implemented as a web service on Azure that ingests video feeds from traffic cameras and outputs the camera calibration parameters. This demo highlights the various stages in the AutoCalib video processing pipeline, and presents two applications: 1) Measurement of on-ground distances between two points in the image and 2) Measurement of vehicle speeds.