David-Traian IANCU, Adina-Magda FLOREA
University Politehnica of Bucharest, 313 Splaiul Independenței, Bucharest, 060042, Romania
firstname.lastname@example.org (*Corresponding author), email@example.com
Abstract: The trajectory prediction task is one of the current challenges regarding computer vision and autonomous driving. However, unlike more common tasks like object detection or semantic segmentation, the problem is far to be resolved and there are still a lot of limitations for the prediction of the trajectory regarding the surrounding vehicles. Only a few models try to make trajectory prediction for a real-life application with live responses and there are only a few datasets to work with, the trajectory being hard to annotate. At least the limitation regarding the dataset annotation can be overcome by using a video generation-based model for trajectory prediction, considering that the video generation task does not require any special annotations. Following a recent study (Iancu et al., 2022), this work proposes some modifications to the PredNet architecture with better results for the trajectory prediction task.
Keywords: Trajectory prediction, Video generation, Autonomous driving, Convolutional neural networks, Long short-term memory networks, PredNet.
>>FULL TEXT: PDF
CITE THIS PAPER AS:
David-Traian IANCU, Adina-Magda FLOREA, An Improved Vehicle Trajectory Prediction Model Based on Video Generation, Studies in Informatics and Control, ISSN 1220-1766, vol. 32(1), pp. 25-36, 2023. https://doi.org/10.24846/v32i1y202303