Sushil Sharma at the VISAPP conference in Rome

Posted on March 01, 2024 

Sushil Sharma, a member of D2iCE, attended the 19th International Conference on Computer Vision Theory and Applications (VISAPP) held in Rome, Italy from February 27 to 29, 2024. During the conference, he discussed the importance of trajectory prediction for vehicle autonomy, emphasizing the challenges posed by the diverse and uncertain behaviors of agents in real-world scenarios. It proposes a learning-based approach, BEVSeg2TP, which integrates perception and trajectory prediction into a unified system to enhance performance. The method employs a surround-view camera bird’s-eye-view perspective, utilizing deep learning techniques for the semantic segmentation of objects and predicting the ego vehicle's trajectory using a spatiotemporal probabilistic network (STPN). By leveraging transformer-based models and cross-attention information, the proposed system achieves improved accuracy in trajectory prediction compared to existing approaches, as demonstrated on the nuScenes dataset, contributing to safer and more effective autonomous driving perception systems.

Full paper details are available here:

BEVSeg2TP: Surround View Camera Bird’s-Eye-View Based Joint Vehicle Segmentation and Ego Vehicle Trajectory Prediction
Authors: Sushil Sharma, Arindam Das, Ganesh Sistu, Mark Halton and Ciaran Eising