Occluded Area Detection Based on Sensor Fusion and Panoptic Segmentation

Detecting occluded areas in a driving environment is crucial to preventing traffic accidents against hidden road agents coming out from such occluded areas. Our previous work proposed a novel detection method that can offer geometric information of the detected areas by utilizing camera and LiDAR sensor fusion. However, it had difficulty identifying individual areas formed by different objects without information about distinct objects. Thus, the objective of this study was to improve our previous methodology, and panoptic segmentation, which can distinguish between individual objects and offer semantic class labels of the object, was adopted to overcome the limitation. Evaluation results revealed that our proposed methodology could achieve satisfactory results in occlusion area detection and superior accuracy in estimating hidden road agent types in the detected areas.