Table 13.

Literature Analysis: 3D Voxel Grid-based 3D Object Recognition Methods.

	SegV Net [63]	SECONDX [64]	LidarMTL [65]
Detector Category	One-stage	Two-stage	Two-stage
Environment	Outdoor	Outdoor	Outdoor
Projection	BEV	FV	BEV
Scenario	Ambiguous vehicles identification scenario from point cloud	Multi class 3D object detection scenario with a single model	Dynamic object detection and static road understanding scenario
Advantage(s)	Encodes the semantic context information in the feature maps to distinguish ambiguous vehicle for better detection	Provides multiple class support in a single model.	Performs robust 3D object recognition in complicated environment Also useful for online localization
Limitation(s)	Partial occlusion leads to false positives	Performance is not satisfactory for all the classes (e.g., cyclist and pedestrian.	The necessity of using loss weights with grid search