Prediction of Pedestrians Intentions
Place: Large Lecture Room
Affiliation: ISIS Research Group, Universidad de Alcalá de Henares, Madrid. Spain.
Advanced Driver Assistance Systems have significantly improved in the latest years, becoming series products in mass-produced commercial vehicles. Pedestrian recognition is a reality long time ago in the automotive industry. In spite of that, OEMs demand systems with higher detection anticipation, capable of increasing the probability to avoid accidents by means of active actuation over the vehicle. The anticipated detection of pedestrian, just 100ms or 200ms in advance, constitutes a major difference in terms of safety. This seminar will present some preliminary ideas to estimate pedestrians intentions in an anticipated manner, with special emphasis on the behavioral transition in the gait cycle. For this purpose, we propose the use of low dimensional latent spaces in which pedestrians movements are more intuitive and predictable.