Velocity estimation of a mobile mapping vehicle using filtered monocular optical flow

• Autor(a) principal: Barbosa, R. L.
• Data de Publicação: 2007
• Outros Autores: Silva, J. F.C. [UNESP], Meneguette, M. [UNESP], Gallis, R. B.A. [UNESP]
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/221067
• topic: Image Orientation; Image Sequence; Mobile Mapping; Monocular Velocity; Optical Flow

Generally, a road image sequence acquired by a mobile mapping system (MMS) is oriented by integrating the Global Positioning System (GPS) and an inertial navigation system (INS) data. Alternatively, this article presents a methodology completely based only on data derived from a road image sequence acquired by a low cost land based MMS as an alternative to orient the images without any auxiliary data so that the derived information comes from the internal image motion through the optical flow. The vehicle velocity computation is based on the monocular optical flow of an image sequence captured by a video camera mounted on the top of a vehicle that travels in a flat urban road without any auxiliary data. With the estimated velocity and the constant image sequence time interval the mobile's relative position can be computed. No matter the technique, the optical flow computation is very sensible to the noise caused by the image acquisition process under real conditions. The noise sources such as high variation of illumination and camera vibration, among others, can affect the velocity estimation if the dense optical flow is used. In order to avoid such drawbacks the translational velocity is computed from a reduced amount of optical flow vectors, exactly those that represent the effective displacement. These vectors are taken in certain portions of the images - the region of interest (ROI) - and they are supposed to be detected by the Canny edge detector algorithm, which means they come from edges and consequently they have intensity variation in the images. The optical flow computation is based on Horn and Schunck method because it is simple and easy to implement. The technique based on the detected vectors reveals a potential to be developed. The best result shows that the estimated velocity is as good as 1% less than the one determined in the control surveying mission. Additionally, the amount of vectors is only 560 instead of 720 x 480 of the dense flow (original image size) or about 230, 000 vectors of a reduced quadrilateral image. These results indicate that the implemented technique contributes to reach a better translational velocity estimation and therefore to the vehicle displacement which lets one know the relative position of the MMS from a road image sequence without any auxiliary data.