纯点云kitti格式自制数据集

纯点云kitti格式数据集制作，参考官网的格式
kitti数据集官网：https://www.cvlibs.net/datasets/kitti/
kitti格式内容如下：（由于是纯点云的数据集，所有有关图像的部分我都设置的默认值，截断程度、遮挡率这些也都设置的0）

数据文件结构如下

my_kitti
├── ImageSets
│   ├── val.txt
│   ├── train.txt
├── testing
│   ├── label_2
│   ├── velodyne
├── training
│   ├── label_2
│   ├── velodyne

数据标注是通过SUSTechPOINTS进行的，关于如何使用SUSTechPOINTS进行标注，请参考我的另一篇文章https://blog.csdn.net/m0_64293675/article/details/144189633?spm=1001.2014.3001.5502。生成的json标注文件部分内容如下：

制作数据集的python脚本如下:

import os 
import numpy as np
import json
import shutil 
import random
from tqdm import tqdm
import open3d as o3d

namesclass={
    'Car': 'Car',
    'Pedestrian': 'Pedestrian',
    'Bicycle': 'Cyclist',
    'ScooterRider':'Cyclist',
    'Truck':'Car',
    'Motorcycle':'Cyclist',
    'Van':'Car',
    'Bus':'Car',
    'TourCar':'Car',
    'Scooter':'Cyclist',
    'Cyclist':'Cyclist'
}

def pcd2bin(pcdfolder, binfolder):
    current_path = os.getcwd()
    ori_path = os.path.join(current_path, pcdfolder)
    file_list = os.listdir(ori_path)
    des_path = os.path.join(current_path, binfolder)
    
    os.makedirs(des_path, exist_ok=True)
    for file in tqdm(file_list): 
        (filename,extension) = os.path.splitext(file)
        velodyne_file = os.path.join(ori_path, filename) + '.pcd'
        pcd = o3d.io.read_point_cloud(velodyne_file)
        # 获取点云数据的NumPy数组
        points = np.asarray(pcd.points)  
        # 反射强度默认设置成0，训练中基本也不会用到
        intensities = np.zeros(points.shape[0])
        # intensities = np.asarray(pcd.colors)[:, 0] if pcd.has_colors() else np.zeros(points.shape[0])

        # 合并点云坐标和强度信息到一个数组中
        point_cloud_data = np.column_stack((points, intensities))
        # print("points shape:   ",point_cloud_data.shape)
        point_cloud_data = point_cloud_data.astype(np.float32)
        velodyne_file_new = os.path.join(des_path, filename) + '.bin'
        point_cloud_data.tofile(velodyne_file_new)


def json2txt(json_paths,txt_paths):
    json_path_list = os.listdir(json_paths)
    os.makedirs(json_paths, exist_ok=True)
    os.makedirs(txt_paths, exist_ok=True)
    
    for json_path in json_path_list:
        # print(json_path)
        (filename,extension) = os.path.splitext(json_path)
        json_path_new = json_paths + "/" +json_path
        txt_file_new = os.path.join(txt_paths, filename) + '.txt'
        print(txt_file_new)
        with open(json_path_new,'r',encoding='utf8')as fp:
            json_data = json.load(fp)
            for i in range(len(json_data)):
                obj_id = json_data[i]["obj_id"]
                obj_type = json_data[i]["obj_type"]
                obj_type_final = namesclass[obj_type]
                
                position_x = json_data[i]["psr"]["position"]["x"]
                position_y = json_data[i]["psr"]["position"]["y"]
                position_z = json_data[i]["psr"]["position"]["z"]
                
                # scale_x y z 分别表示长宽高
                # 数据集要求依次输入高宽长
                scale_x = json_data[i]["psr"]["scale"]["x"]
                scale_y = json_data[i]["psr"]["scale"]["y"]
                scale_z = json_data[i]["psr"]["scale"]["z"]
                #  标注文件中偶尔会有长宽高为负数的情况，需要取绝对值
                if scale_x<0 or scale_y<0 or scale_z<0:
                    scale_x = abs(scale_x)
                    scale_y = abs(scale_y)
                    scale_z = abs(scale_z)
                    
                # sustechoint软件标注的坐标是3D-BOX的中心坐标，kitti数据集的格式要求的是底面中心坐标
                position_z_new = position_z - 0.5*scale_z
                
                rotation_x = json_data[i]["psr"]["rotation"]["x"]
                rotation_y = json_data[i]["psr"]["rotation"]["y"]
                rotation_z = json_data[i]["psr"]["rotation"]["z"]
                line = obj_type_final +  " " + "0"+ " " + "0" + " " + "0" + " " "100" +" " +"100"+ " " +"200"+" " +"200" +" " + str(scale_z) + " " + str(scale_y) + " " + str(scale_x) +" " + str(position_x) + " " + str(position_y) + " " + str(position_z_new) + " " + str(rotation_z) +"\n"
                with open(txt_file_new,"a") as f:
                    f.write(line)

           
def gen_kitti_data(SrcbinPath,SrctxtPath,kitti_data_path,test_ratio):

    files = os.listdir(SrctxtPath)
    # print(files)
    # 随机打乱文件顺序
    random.shuffle(files)
    # 计算划分数据集的索引
    total_files = len(files)
    train_split = int((1-test_ratio) * total_files)
    val_split = int(test_ratio * total_files)
    

    for file in tqdm(files[:train_split], desc=f'Copying train data'):
        # 复制bin
        src_bin = os.path.join(SrcbinPath, file[:-4] + '.bin')
        dst_bin = os.path.join(kitti_data_path, 'training','velodyne')
        
        # 复制txt和bin
        src_txt = os.path.join(SrctxtPath, file)
        dst_txt = os.path.join(kitti_data_path, 'training','label_2')
        os.makedirs(dst_bin, exist_ok=True)
        shutil.copy(src_bin, os.path.join(dst_bin, file[:-4] + '.bin'))
        os.makedirs(dst_txt, exist_ok=True)
        shutil.copy(src_txt, os.path.join(dst_txt, file))
        
    os.makedirs(kitti_data_path + "/ImageSets", exist_ok=True)
    train_path = kitti_data_path + "/ImageSets" +  "/train.txt"
    file_list = os.listdir(dst_bin)
    data = []
    for file in file_list: 
        (filename,extension) = os.path.splitext(file)
        data.append(filename)
    for i in data:
        with open(train_path,"a") as f:
            f.write(str(i) + "\n")
    for file in tqdm(files[train_split:train_split + val_split], desc=f'Copying validation data'):
         # 复制bin
        src_bin = os.path.join(SrcbinPath, file[:-4] + '.bin')
        dst_bin = os.path.join(kitti_data_path, 'testing','velodyne')
        os.makedirs(dst_bin, exist_ok=True)
        shutil.copy(src_bin, os.path.join(dst_bin, file[:-4] + '.bin'))
        # 同时复制txt
        src_txt = os.path.join(SrctxtPath, file)
        dst_txt = os.path.join(kitti_data_path, 'testing','label_2')
        os.makedirs(dst_txt, exist_ok=True)
        shutil.copy(src_txt, os.path.join(dst_txt, file))
    os.makedirs(kitti_data_path + "/ImageSets", exist_ok=True)
    train_path = kitti_data_path + "/ImageSets" +  "/val.txt"
    file_list = os.listdir(dst_bin)
    data = []
    for file in file_list: 
        (filename,extension) = os.path.splitext(file)
        data.append(filename)
    for i in data:
        with open(train_path,"a") as f:
            f.write(str(i) + "\n")
    print("数据集划分完成！")



# 1、pcd转bin   
pcd2bin("lidar","velodyne")    
        
# 2、json格式的标注文件转txt
json2txt("label_json","label_txt")

# 3、生成数据集
gen_kitti_data("velodyne","label_txt","kitti_dataset",0.1)