Skip to content
Commits on Source (3)
Hide whitespace changes
Inline Side-by-side
README.md
View file @ 8d04e48b
...@@ -30,10 +30,9 @@ graph LR; ...@@ -30,10 +30,9 @@ graph LR;
A[VSLAM]-->|KeyFrames| B[YOLOv4]; A[VSLAM]-->|KeyFrames| B[YOLOv4];
A-->|Pointcloud| C[Object Tracking]; A-->|Pointcloud| C[Object Tracking];
A-->|Camera Pose| C; A-->|Camera Pose| C;
A-->|Pointcloud| E
B-->|Object Detection|C; B-->|Object Detection|C;
C-->|Objects| D[Database]; C-->|3D Objects| D[Database];
D-->|Objects| E[GUI]; D-->|3D Objects| E[GUI];
``` ```
- **Camera/Video**: We use prerecorded videos as our examples and tests. We use two datasets, both augmented with depth perception data, which allows us to reconstruct a dense map for the scene captured from the footage. We also have a mechanism of collecting a monocular video feed (either from a drone or from regular phone camera) and we have Java scripts to break the footage into frames, which we can then perform monocular VSLAM on. Note that monocular VSLAM is not as effective as RGBD VSLAM, as we are unable to recreate the dense color map of the scene, therefore, we limit our functionality with this data source to simply demonstrating our ability to perform VSLAM on footage collected from an arbitrary camera source. - **Camera/Video**: We use prerecorded videos as our examples and tests. We use two datasets, both augmented with depth perception data, which allows us to reconstruct a dense map for the scene captured from the footage. We also have a mechanism of collecting a monocular video feed (either from a drone or from regular phone camera) and we have Java scripts to break the footage into frames, which we can then perform monocular VSLAM on. Note that monocular VSLAM is not as effective as RGBD VSLAM, as we are unable to recreate the dense color map of the scene, therefore, we limit our functionality with this data source to simply demonstrating our ability to perform VSLAM on footage collected from an arbitrary camera source.
......