该数据库是目前规模较大的行人数据库，采用车载摄像头拍摄，约10个小时左右，视频的分辨率为640x480，30帧/秒。标注了约249,884帧（约137分钟），350000个矩形框，2300个行人，另外还对矩形框之间的时间对应关系及其遮挡的情况进行标注。

下载好的Caltech行人数据集，解压放在Caltech文件夹内，目录情况如下图。

1 图片转换

将.seq文件转换为.jpg文件并保存在某一目录下。

seq2jpg.py

# -*- coding:utf-8 -*-
import os
import fnmatch


def open_save(file, savepath):
    f = open(file, 'rb')
    # 将seq文件的内容转化成str类型
    string = f.read().decode('latin-1')

    # splitstring是图片的前缀，可以理解成seq是以splitstring为分隔的多个jpg合成的文件
    splitstring = "\xFF\xD8\xFF\xE0\x00\x10\x4A\x46\x49\x46"  
    # split函数做一个测试,因此返回结果的第一个是在seq文件中是空，因此后面省略掉第一个
    strlist = string.split(splitstring)

    f.close()
    count = 0  

    # 遍历每一个jpg文件内容，然后加上前缀合成图片
    for img in strlist:        
        filename = str(count)+'.jpg'
        filename = savepath + filename
        if count > 0:                                    
            i = open(filename, 'wb+')
            i.write(splitstring.encode('latin-1'))
            i.write(img.encode('latin-1'))
            i.close()
        count = count + 1
        print("Generating JPEGImages jpg file of picture:{}".format(filename))


def main():
    seq_inputdir = "E:\\dataset\\Caltech"
    jpg_outputdir = "E:\\dataset\\Caltech_VOC\\JPEGImages"

    for dir_set in os.listdir(seq_inputdir):
        # 若果是set01、set02……这类文件夹
        if dir_set[0:3] == "set" and os.path.splitext(dir_set)[1] == '':
            dir_set_seq = os.path.join(seq_inputdir, dir_set)
            for dir_seq in os.listdir(dir_set_seq):
                if fnmatch.fnmatch(dir_seq, '*.seq'):
                    file_dir = os.path.join(dir_set_seq, dir_seq)
                    save_dir = os.path.join(jpg_outputdir, dir_set) + '_' + dir_seq.split('.')[0] + '_'
                    open_save(file_dir, save_dir)


if __name__ == '__main__':
    main()
    print("Success！")

2 文本转换

将.vbb文件转换为.xml文件并保存在某一目录下。

vbb2voc.py

# -*- coding:utf-8 -*-
# -*- coding:utf-8 -*-
import os, glob
from scipy.io import loadmat
from collections import defaultdict
import numpy as np
from lxml import etree, objectify


def vbb_anno2dict(vbb_file, cam_id):
    filename = os.path.splitext(os.path.basename(vbb_file))[0]

    # 定义字典对象annos
    annos = defaultdict(dict)
    vbb = loadmat(vbb_file)
    # object info in each frame: id, pos, occlusion, lock, posv
    objLists = vbb['A'][0][0][1][0]
    objLbl = [str(v[0]) for v in vbb['A'][0][0][4][0]]     # 可查看所有类别
    # person index
    person_index_list = np.where(np.array(objLbl) == "person")[0]   # 只选取类别为‘person’的xml
    for frame_id, obj in enumerate(objLists):
        if len(obj) > 0:
            frame_name = str(cam_id) + "_" + str(filename) + "_" + str(frame_id+1) + ".jpg"
            annos[frame_name] = defaultdict(list)
            annos[frame_name]["id"] = frame_name
            annos[frame_name]["label"] = "person"
            for id, pos, occl in zip(obj['id'][0], obj['pos'][0], obj['occl'][0]):
                id = int(id[0][0]) - 1  # for matlab start from 1 not 0
                if not id in person_index_list:  # only use bbox whose label is person
                    continue
                pos = pos[0].tolist()
                occl = int(occl[0][0])
                annos[frame_name]["occlusion"].append(occl)
                annos[frame_name]["bbox"].append(pos)
            if not annos[frame_name]["bbox"]:
                del annos[frame_name]
    print(annos)
    return annos


def instance2xml_base(anno, bbox_type='xyxy'):
    """bbox_type: xyxy (xmin, ymin, xmax, ymax); xywh (xmin, ymin, width, height)"""
    assert bbox_type in ['xyxy', 'xywh']
    E = objectify.ElementMaker(annotate=False)
    anno_tree = E.annotation(
        E.folder('VOC2014_instance/person'),
        E.filename(anno['id']),
        E.source(
            E.database('Caltech pedestrian'),
            E.annotation('Caltech pedestrian'),
            E.image('Caltech pedestrian'),
            E.url('None')
        ),
        E.size(
            E.width(640),
            E.height(480),
            E.depth(3)
        ),
        E.segmented(0),
    )
    for index, bbox in enumerate(anno['bbox']):
        bbox = [float(x) for x in bbox]
        if bbox_type == 'xyxy':
            xmin, ymin, w, h = bbox
            xmax = xmin+w
            ymax = ymin+h
        else:
            xmin, ymin, xmax, ymax = bbox
        E = objectify.ElementMaker(annotate=False)
        anno_tree.append(
            E.object(
            E.name(anno['label']),
            E.bndbox(
                E.xmin(int(xmin)),
                E.ymin(int(ymin)),
                E.xmax(int(xmax)),
                E.ymax(int(ymax))
            ),
            E.difficult(0),
            E.occlusion(anno["occlusion"][index])
            )
        )
    return anno_tree


def parse_anno_file(vbb_inputdir, vbb_outputdir):
    # annotation sub-directories in hda annotation input directory
    assert os.path.exists(vbb_inputdir)
    sub_dirs = os.listdir(vbb_inputdir)     # 对应set00,set01...
    for sub_dir in sub_dirs:
        print("Parsing annotations of camera: ", sub_dir)
        cam_id = sub_dir
        # 获取某一个子set下面的所有vbb文件
        vbb_files = glob.glob(os.path.join(vbb_inputdir, sub_dir, "*.vbb"))
        for vbb_file in vbb_files:
            # 返回一个vbb文件中所有的帧的标注结果
            annos = vbb_anno2dict(vbb_file, cam_id)
            
            if annos:
                # 组成xml文件的存储文件夹，形如“/Users/chenguanghao/Desktop/Caltech/xmlresult/”
                vbb_outdir = vbb_outputdir
                                              
                # 如果不存在 vbb_outdir
                if not os.path.exists(vbb_outdir):
                    os.makedirs(vbb_outdir)

                for filename, anno in sorted(annos.items(), key=lambda x: x[0]):                  
                    if "bbox" in anno:
                        anno_tree = instance2xml_base(anno)
                        outfile = os.path.join(vbb_outdir, os.path.splitext(filename)[0]+".xml")
                        print("Generating annotation xml file of picture: ", filename)
                        # 生成最终的xml文件，对应一张图片
                        etree.ElementTree(anno_tree).write(outfile, pretty_print=True)


def visualize_bbox(xml_file, img_file):
    import cv2
    tree = etree.parse(xml_file)
    # load image
    image = cv2.imread(img_file)
    origin = cv2.imread(img_file)
    # 获取一张图片的所有bbox
    for bbox in tree.xpath('//bndbox'):
        coord = []
        for corner in bbox.getchildren():
            coord.append(int(float(corner.text)))
        print(coord)
        cv2.rectangle(image, (coord[0], coord[1]), (coord[2], coord[3]), (0, 0, 255), 2)
    # visualize image
    cv2.imshow("test", image)
    cv2.imshow('origin', origin)
    cv2.waitKey(0)


def main():
    vbb_inputdir = "F:\\dataset\\Caltech\\annotations"
    vbb_outputdir = "F:\\dataset\\Caltech_VOC\\Annotations"
    parse_anno_file(vbb_inputdir, vbb_outputdir)


if __name__ == "__main__":
    main()
    print("Success!")

得到如下文件：

Annotations内包含122,187个xml文件。（每一个xml文件内都包含“person”标签）
JPEGImages内包含249,884个jpg文件。（很多jpg文件内不含有目标，所以要比xml文件多）

3 数据筛选

由于Annotations内包含122,187个xml文件，JPEGImages内包含249,884个jpg文件，不能实现一一对应，所以要去除多余的jpg文件。

但，其实很多xml文件内标注的person由于目标太小，导致及其不清晰，应该将其去除，再进一步处理。

这是数据集中的一张图片，猜猜人在哪？
欸，在这里！！


由于这些目标让我难以描述，所以将此类目标（尺寸较小）去除掉。

3.1 根据目标面积进行筛选

将符合条件的目标保存到另一个目录下。

# -*- coding:utf-8 -*-
import xml.etree.ElementTree as ET
import os

space = 40*70
xml_path = "F:\\dataset\\Caltech_VOC\\Annotations"
save_path = "F:\\dataset\\Caltech_VOC\\Annotations_"


def filter_xml():
    for file_name in os.listdir(xml_path):
        file_path = os.path.join(xml_path, file_name)
        print(file_path)
        in_file = open(file_path)
        tree = ET.parse(in_file)  # ET是一个xml文件解析库，ET.parse（）打开xml文件。parse--"解析"
        root = tree.getroot()  # 获取根节点

        write_flag = False

        for obj in root.findall('object'):  # 找到根节点下所有“object”节点
            bbox = obj.find('bndbox')
            x1 = int(bbox.find('xmin').text)
            y1 = int(bbox.find('ymin').text)
            x2 = int(bbox.find('xmax').text)
            y2 = int(bbox.find('ymax').text)

            if (x2-x1)*(y2-y1) >= space:
                write_flag = True
        if write_flag:
            tree.write(os.path.join(save_path, file_name))
            print("save file: {}".format(os.path.join(save_path, file_name)))


if __name__ == '__main__':
    filter_xml()

3.2 根据xml保存jpg

将含有xml文件的jpg保存在另一目录下。

import os
import shutil
from tqdm import tqdm

xml_path = "F:\\dataset\\Caltech_VOC\\Annotations_"
jpg_path = "F:\\dataset\\Caltech_VOC\\JPEGImages"
save_path = "F:\\dataset\\Caltech_VOC\\JPEGImages_"


if __name__ == '__main__':
    filter = []
    print("filter loading...")
    for file in tqdm(os.listdir(xml_path)):
        filter.append(os.path.splitext(file)[0])

    for file in os.listdir(jpg_path):
        if os.path.splitext(file)[0] in filter:
            old_path = os.path.join(jpg_path, file)
            new_path = os.path.join(save_path, file)
            shutil.copyfile(old_path, new_path)
            print("Success save jpg file: {}".format(new_path))

    print("Success!")

处理完成后：