KITTI自动驾驶数据集使用与点云可视化实例

KITTI数据集

KITTI数据集是自动驾驶领域最著名的公开数据集之一，由德国卡尔斯鲁厄理工学院和丰田美国技术研究院联合创建。它包含：

超过6小时的交通场景数据
高分辨率RGB图像(左/右相机)
3D激光雷达点云数据(64线Velodyne激光扫描仪)
3D物体标注(汽车、行人、自行车等)
立体视觉、光流、视觉里程计等多种任务数据

数据集下载与结构

KITTI数据集可从官网下载：http://www.cvlibs.net/datasets/kitti/

主要目录结构：

kitti/
├── raw_data/           # 原始数据
│   ├── 2011_09_26/     # 按日期组织的采集数据
│   └── ...
├── object/             # 3D物体检测任务数据
│   ├── training/
│   │   ├── image_2/    # 左相机图像
│   │   ├── velodyne/   # 点云数据(.bin)
│   │   └── label_2/    # 标注文件
│   └── testing/
└── tracking/           # 多目标跟踪数据

点云数据可视化实例

1. 使用Python进行点云可视化

import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

def read_velodyne_bin(bin_path):
    """读取KITTI的.bin点云文件"""
    point_cloud = np.fromfile(bin_path, dtype=np.float32)
    return point_cloud.reshape((-1, 4))  # x,y,z,反射率

def visualize_point_cloud(points):
    """3D可视化点云"""
    fig = plt.figure(figsize=(10, 8))
    ax = fig.add_subplot(111, projection='3d')
    
    # 只取前3列(x,y,z)，忽略反射率
    ax.scatter(points[:, 0], points[:, 1], points[:, 2], 
               s=1, c=points[:, 2], cmap='viridis')
    
    ax.set_xlabel('X (m)')
    ax.set_ylabel('Y (m)')
    ax.set_zlabel('Z (m)')
    ax.set_title('KITTI点云可视化')
    plt.show()

# 示例使用
bin_file = "kitti/object/training/velodyne/000000.bin"
points = read_velodyne_bin(bin_file)
visualize_point_cloud(points)

2. 使用Open3D进行更高级的可视化

import open3d as o3d

def visualize_with_open3d(points):
    """使用Open3D可视化点云"""
    pcd = o3d.geometry.PointCloud()
    pcd.points = o3d.utility.Vector3dVector(points[:, :3])  # 只取x,y,z
    
    # 设置颜色基于Z轴高度
    colors = plt.get_cmap("viridis")((points[:, 2] - points[:, 2].min()) / 
                                    (points[:, 2].max() - points[:, 2].min()))
    pcd.colors = o3d.utility.Vector3dVector(colors[:, :3])
    
    # 可视化
    o3d.visualization.draw_geometries([pcd],
                                      window_name="KITTI点云",
                                      width=800,
                                      height=600)

# 使用示例
visualize_with_open3d(points)

3. 结合图像和点云的可视化

import cv2
from PIL import Image

def project_points_to_image(points, calib_path, image_path):
    """将点云投影到图像平面"""
    # 读取标定数据
    with open(calib_path, 'r') as f:
        lines = f.readlines()
        P2 = np.array(lines[2].strip().split(' ')[1:], dtype=np.float32).reshape(3, 4)
        R0_rect = np.array(lines[4].strip().split(' ')[1:], dtype=np.float32).reshape(3, 3)
        Tr_velo_to_cam = np.array(lines[5].strip().split(' ')[1:], dtype=np.float32).reshape(3, 4)
    
    # 转换点云到相机坐标系
    points_3d = points[:, :3]
    points_3d_hom = np.hstack([points_3d, np.ones((points_3d.shape[0], 1))])
    points_cam = R0_rect @ (Tr_velo_to_cam @ points_3d_hom.T)
    points_cam = points_cam.T
    
    # 投影到图像平面
    points_img = P2 @ np.hstack([points_cam[:, :3], np.ones((points_cam.shape[0], 1))]).T
    points_img = points_img.T
    points_img[:, :2] /= points_img[:, 2][:, np.newaxis]
    
    # 过滤在图像外的点
    img = Image.open(image_path)
    width, height = img.size
    mask = (points_img[:, 0] >= 0) & (points_img[:, 0] < width) & \
           (points_img[:, 1] >= 0) & (points_img[:, 1] < height) & \
           (points_img[:, 2] > 0)
    points_img = points_img[mask]
    points_3d = points_3d[mask]
    
    # 可视化
    img = np.array(img)
    fig, ax = plt.subplots(1, 1, figsize=(12, 5))
    ax.imshow(img)
    
    # 根据深度着色
    depths = points_img[:, 2]
    colors = plt.cm.viridis((depths - depths.min()) / (depths.max() - depths.min()))
    
    ax.scatter(points_img[:, 0], points_img[:, 1], 
               c=colors, s=5, alpha=0.7)
    plt.title("点云投影到图像")
    plt.show()

# 使用示例
calib_file = "kitti/object/training/calib/000000.txt"
image_file = "kitti/object/training/image_2/000000.png"
project_points_to_image(points, calib_file, image_file)