Page 38 - Read Online

Chen et al. Intell Robot 2024;4:179-95 I http://dx.doi.org/10.20517/ir.2024.11 Page 187

Table 1. Parameters of the depth camera
Camera parameters
Size 71.9 mm × 19.2 mm × 10.6 mm
Resolution 224 × 172 pixels
Frame rate Up to 60 fps
Average power consumption 300 mW
Measurement range 0.1 ∼ 4 m
Weight 13 g

Figure 4. The experiment setup for evaluating the proposed staircase feature extraction method. IMU: Inertial measurement unit.

in this paper, contributes to reducing absolute trajectory errors in seven out of eight trials. This implies that
the camera’s estimated motion trajectories, obtained through the proposed method, align more closely with
the ground truth than trajectories derived solely from extracting convex corner points as features.

To demonstrate that our feature extraction system is more suitable for visually constrained prosthetic systems,
we conducted comparative experiments with four repeated trials using the Open3D built-in ICP algorithm [26]
to obtain relative displacement. In the experiment, to ensure fairness, both our algorithm and Open3D’s al-
gorithm used the same 2D point cloud. The point cloud was replicated in columns to transform into a 3D
point cloud for calling Open3D’s built-in ICP method. Based on statistics, the average processing time for
Open3D ICP algorithm on the replicated five-column point cloud is ∼10 ms, longer than the simplified KNN-
ICP method within the 2D plane (3 ms). The odometer trajectory estimation and absolute errors for different
algorithms are shown in Figure 6. The comparison results on the absolute trajectory error and the processing
time demonstrate that the proposed method can achieve better estimation accuracy and a faster processing
speed due to the dimension reduction process.

3.2 Evaluation of the robustness of the proposed method
To evaluate the robustness of our approach across wider scenarios, we tested it on another staircase with stairs
that are 27.5 cm wide and 14.5 cm high. The absolute trajectory error results in four repeated trials on this
higher staircase are presented in Figure 7. The results show that the absolute trajectory errors on both kinds
of stairs are of the same scale, demonstrating the robustness of the proposed method on various stair sizes.