Variational Voxel Pseudo Image Tracking
This work addresses uncertainty estimation for critical applications like robotics and autonomous driving, but it is incremental as it builds on existing Voxel Pseudo Image Tracking methods.
The paper tackles 3D single object tracking by proposing a Variational Neural Network-based method that estimates uncertainties in features, improving tracking performance with uncertainty-aware cross-correlation modules.
Uncertainty estimation is an important task for critical problems, such as robotics and autonomous driving, because it allows creating statistically better perception models and signaling the model's certainty in its predictions to the decision method or a human supervisor. In this paper, we propose a Variational Neural Network-based version of a Voxel Pseudo Image Tracking (VPIT) method for 3D Single Object Tracking. The Variational Feature Generation Network of the proposed Variational VPIT computes features for target and search regions and the corresponding uncertainties, which are later combined using an uncertainty-aware cross-correlation module in one of two ways: by computing similarity between the corresponding uncertainties and adding it to the regular cross-correlation values, or by penalizing the uncertain feature channels to increase influence of the certain features. In experiments, we show that both methods improve tracking performance, while penalization of uncertain features provides the best uncertainty quality.