bouding box does not match to object

[nghiabka]

Thank you for your excellent work.

I am trying to draw ground truth with point cloud data. However, the bounding box does not match the object. I am so confused. I would be appreciate any comments or suggestions. Thank you.

Here is my code to visualize box

import open3d as o3d
import open3d
import numpy as np
import pandas as pd
import math

from scipy.spatial.transform import Rotation

def load_point_cloud(pcd_path):
    """
    Load the point cloud from a .pcd file.
    """
    pcd = o3d.io.read_point_cloud(pcd_path)
    return pcd

def parse_labels(label_path):
    """
    Parse label data from the given label file.

    Each line in the label file should have the following format:
    Class name, Track ID, Height (h), Length (l), Width (w), X, Y, Z, rotation (rot)

    Returns:
        List of dictionaries with bounding box parameters.
    """
    labels = []
    with open(label_path, 'r') as file:
        for line in file.readlines():
            print("line ", line)
            data = line.strip().split(' ')
            # print("data ", data)
            if len(data) == 9:
                label = {
                    'class_name': data[0].strip(),
                    'track_id': data[1].strip(),
                    'h': float(data[2].strip()),
                    'l': float(data[3].strip()),
                    'w': float(data[4].strip()),
                    'x': float(data[5].strip()),
                    'y': float(data[6].strip()),
                    'z': float(data[7].strip()),
                    'rot': float(data[8].strip())  # Rotation in radians
                }
                labels.append(label)
    return labels

pcd_path = "/data/Sit/Cafe_street/Cafe_street_1/velo/concat/data/100.pcd"
label_path = "/data/Sit/Cafe_street/Cafe_street_1/label_3d/100.txt"

point_data = load_point_cloud(pcd_path)
bounding_boxes = parse_labels(label_path)
print(len(bounding_boxes))
vis_elements = [point_data]  # List of all visual elements to add to Open3D visualizer

vis = open3d.visualization.Visualizer()
vis.create_window()

vis.get_render_option().point_size = 4.0
vis.get_render_option().background_color = np.ones(3) * 0.25

axis_pcd = open3d.geometry.TriangleMesh.create_coordinate_frame(
            size=1.0, origin=[0, 0, 0])
vis.add_geometry(axis_pcd)

axis_pcd = open3d.geometry.TriangleMesh.create_coordinate_frame(
size=1.0, origin=[0, 0, 0])
vis.add_geometry(axis_pcd)

pts = open3d.geometry.PointCloud()

point_data = np.asarray(point_data.points)

print(point_data.shape)

for gt in bounding_boxes:
    if gt['class_name']=="Pedestrian":
        print(gt['rot'])
        rot_mat = Rotation.from_rotvec([0, 0, gt['rot']]).as_matrix()
        bb_gt = open3d.geometry.OrientedBoundingBox([gt['x'], gt['y'], gt['z']], rot_mat, [gt['w'], gt['l'], gt['h']])
        bb_gt.color = (0.0, 1.0, 2.0)
        vis.add_geometry(bb_gt)

pts.points = open3d.utility.Vector3dVector(point_data[:, :3])

view_control = vis.get_view_control()
view_control.set_front([0, 0, -1])
view_control.set_lookat([1, 2, 3])  # Adjust the look-at position if needed
view_control.set_up([0, -1, 0])
view_control.set_zoom(0.8)

vis.add_geometry(pts)

vis.run()

[nghiabka]

I found a solution. We need to combine the label and trajectory information to get the final bounding box.