Högskolan i Skövde

Recognizing Pedestrians intention to cross a street and predicting their imminent crossing action are major challenges for advanced driver assistance systems (ADAS) and Autonomous Vehicles (AV). In this paper we address these problems by proposing a new neural network architecture that uses feature fusion. The approach is used to recog[1]nise/predict the pedestrians intention/action 1.5 sec (45 frames) ahead. We evaluate our approach on the recently suggested benchmark by Rasouli et al. and show that our approach outperforms current state of the art models. We observe further improved results when the model is trained and tested on a stronger balanced subset of the PIE dataset.