Table 1.
Summary of 3D Object Detection Datasets
| Types of Datasets | Sensors | 3D Boxes | Classes | Number of Scenes | Annotated Frames |
|---|---|---|---|---|---|
| Waymo | RGB+Li DAR | 12M | 4 | 1k | 200k |
| ApolloSc ape | RGB+Li DAR | 70K | 35 | Nil | 140K |
| nuScene s | RGB+Li DAR | 1.4 M | 23 | 1k | 40 k |
| A*3D | RGB+Li DAR | 230 K | 7 | Nil | 39k |
| H3D | RGB+Li DAR | 1.1 M | 8 | 160 | 27k |
| Lyft Level 5 | RGB+Li DAR | 1.3 M | 9 | 366 | 46k |
| KITTI | RGB+Li DAR | 200 K | 8 | 22 | 15k |
Figure 1.
Architecture of YOLOv4 [20]
Figure 2.
Architecture of Object Detectors
Table 2.
Comparison of Proposed Model with Existing Models [25]
| Architectures | Input | Output | Metrics | Outcome |
|---|---|---|---|---|
| YOLOv3 WCCS | RGB frame | Speed Detect | MAE MAP | 78% 85% |
| YOLOV3 V3-Tiny | Speed RGB frame Position | Speed | Successful Episodes | 84% |
| YOLOv3 FZ-NMS | Speed RGB frame IMU | Detect Speed | MAP AP | 89% 93% |
| Proposed | RGB frame LIDAR | Speed Detect | AP MAP | 98% 95% |
Table 3.
Comparison of Average Precision of Proposed Approach with Conventional Algorithms [26]
| Models | AP (Average Precision) | Computational Time (ms) |
|---|---|---|
| Tiny YOLOv3 | 0.40 | 10 |
| Late Fusion | 0.40 | 14 |
| RVNet | 0.56 | 17 |
| Proposed | 0.74 | 14.85 |