Dense Scene Flow Estimation with the GVF Snake Model for Obstacle Detection Using a 3D Sensor in the Path-Planning Module

Sobers Francis; Sreenatha Anavatti; Mathew Garratt; Osama Hassan; Shabaan Ali

doi:10.2478/ijssis-2023-0015

.blurhash-client-img { display: none !important; }

Dense Scene Flow Estimation with the GVF Snake Model for Obstacle Detection Using a 3D Sensor in the Path-Planning Module

International Journal on Smart Sensing and Intelligent Systems

Volume 16 (2023): Issue 1 (January 2023)

By: Sobers Francis, Sreenatha Anavatti, Mathew Garratt, Osama Hassan and Shabaan Ali

Open Access

|Nov 2023

Figures & Tables

System Overview. (a) PMD camera mounted on P3DX. (b) Microsoft Kinect mounted on the AscTec pelican quadrotor. P3DX, Pioneer 3DX.

PMD Depth Image Analysis with Barrel Distortions. The distance between the PMD camera and the fixed board are: (a) 500 mm (b) 750 mm (c) 1000 mm (d) 1250 mm (e) 1500 mm (f) 1750 mm (g) 2000 mm (h) 2250 mm (i) 2500 mm.

Middlebury datasets: (a) Rubber Whale (b) Dimetrodon (c) Urban2.

Error Plots for Middlebury Datasets (a) Rubberwhale Dataset (b) Dimetrodon Dataset (c) Urban2 Dataset.

Dense Scene flow in Various Directions (Bright Light); the Coordinate on the Bottom Left is the Camera Coordinate. (a) The x-flow and y-flow components are horizontal and vertical, respectively. To conduct this experiment, a 45º translation of the paperboard's x and y coordinates was made in the direction of the camera plane. (b) The z-flow component is horizontal, and the x-flow component is vertical. For this experiment, a combined motion in y and z coordinates was used to translate the paperboard in the direction of the camera plane. (c) The y-flow component is horizontal, and the z-flow component is vertical. To conduct this experiment, a combined motion in x and z coordinates was applied to the paperboard to translate it toward the camera plane.

The Camera Coordinate Is in the Lower Left Corner of a Dense Scene Flow in Various Directions with Partial Light. (a) The x-flow and y-flow components are horizontal and vertical, respectively. The paperboard was moved toward the camera plane while conducting this experiment with a 45º motion in the x and y coordinates. (b) The horizontal component is z-flow and the vertical component is x-flow. Doing this experiment, the paperboard was translated toward the camera plane with a combined motion in y and z coordinates. (c) Z-flow is the horizontal component, and x-flow is the vertical component. For this experiment, a combined motion in y and z coordinates was used to translate the paperboard in the direction of the camera plane.

Flow Vectors Are Shown with Close Objects Shown in Red and Far Objects Shown in Blue Using Various Methods. (a) Depth frame 1 with color coding. (b) Depth frame 2 with color coding. (c) Moving Obstacle Segmentation. (d) LK. (e) HS. (f) Combined Lk-HS. (g) Combined Lk-HS with GVF. GVF, gradient vector field; Lk-HS, Lucas–Kanade; Horn–Schunck.

Experiment I: (a–c) Range Images with 200 × 200 Pixels; (d–f) Grayscale Images.

Experiment I: 3D Renderings of PMD Camera on GLFW Window. (a) GLFW Frame 1.(b) GLFW Frame 2. (c) GLFW Frame 3.

Different Frames of a Moving Obstacle. (A–J) Frame 1 to Frame 10.

Obstacle Segmentation: Dense Flow Vectors in 3D Plot.

Experiment II: Path Planning with Two Irregular Shaped Obstacles. (A–E) Frame 1 to Frame 5.

Experiment II: Real-Time Rendering of PMD Range Data Using GLFW. (a) Range intensity: Tripod (far). (b) Range intensity: Tripod (near). (c) Range intensity: P3DX(2) (far). (d) Range intensity: P3DX(2) (near). (e) 3D point cloud of obstacle: tripod. (f) 3D point cloud of obstacle: P3DX(2). (g) Experiment II: plot of AGV's X-coordinates vs Y-coordinates. P3DX, Pioneer 3DX.

Experiment III on Moving Obstacle: 3D Renderings of PMD Camera.

Errors for Different Methods Compared with Middlebury Data Using MATLAB_ [Regularization Parameter in GVF (μ = _01), Weighting Term (α = _01)]

(μ = 0.01)(α = 0.01)	AAE	STDAE	EPE	AAE	STDAE	EPE	AAE	STDAE	EPE
Lukas–Kanade	0.3128	0.25895	0.596	0.6691	0.3653	1.535	0.8884	0.5021	8.000
Horn–Schunck	0.3151	0.24985	0.603	0.7141	0.3602	1.625	0.9601	0.6128	8.038
LK–HS	0.3059	0.2625	0.586	0.6632	0.5202	1.627	0.8873	0.7030	8.064
LK–HS–GVF	0.3006	0.2585	0.576	0.6291	0.3086	1.526	0.8765	0.5381	7.870

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/ijssis-2023-0015 | Journal eISSN: 1178-5608

Journal RSS Feed

Language: English

Submitted on: May 1, 2023

Published on: Nov 20, 2023

Published by: Macquarie University, Australia

In partnership with: Paradigm Publishing Services

Publication frequency: 1 issue per year

Keywords:

Related subjects:

Introductions and overviews,

Engineering, other

© 2023 Sobers Francis, Sreenatha Anavatti, Mathew Garratt, Osama Hassan, Shabaan Ali, published by Macquarie University, Australia
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Volume 16 (2023): Issue 1 (January 2023)

Dense Scene Flow Estimation with the GVF Snake Model for Obstacle Detection Using a 3D Sensor in the Path-Planning Module

Figures & Tables

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Figure 8:

Figure 9:

Figure 10:

Figure 11:

Figure 12:

Figure 13:

Figure 14:

Figure 15:

Figure 16:

Errors for Different Methods Compared with Middlebury Data Using MATLAB_ [Regularization Parameter in GVF (μ = _01), Weighting Term (α = _01)]

Paradigm

My account