This paper deals with the estimation of both motorcycle lateral dynamics and road geometry reconstruction. A linear parameter varying (LPV) unknown input observer is designed to estimate the whole motorcycle dynamic states including road banking angles and the rider's steering torque taken into account the variation of the forward velocity. The road bank angle and the lateral slip angle are relevant parameters for improving rider's safety and handling, thus, it is interesting to estimate the road geometry. The observer convergence study is based on Lyapunov theory and the established convergence conditions are expressed in linear matrix inequalities (LMIs) formalism. The main idea consists in getting a set of conditions to design an observer transformed into a polytopic form, which estimates a part of the motorcycle dynamics states independently of some inputs (rider torque) and/or other states (zeros dynamics: roll angle) taken into account the variation of the longitudinal velocity.