Ego Vehicle Speed Estimation Using Optical Flow Analysis

Speed estimation is a burgeoning area of research in the domain of extracting motion information from a sequence of image frames which is commonly known as visual odometry and is an essential task in autonomous vehicles. Despite having dedicated odometers, occlusions, noises, and interferences in a complex urban environment suggest visual odometry as a viable extra option. In this work, we propose an image processing pipeline that enhances speed estimation for an ego-vehicle through optical flow analysis from image pairs captured by a dashboard-mounted camera. This is achieved through post-processing of the optical flow output using the proposed pipeline and justified using a framework proposed along with this work. We have compared speed estimations considering two optical flow estimation techniques, post-processing techniques on estimated optical flow, and two different speed estimation approaches which altogether demonstrate effective modifications that can enhance speed estimation using optical flow.