深度学习与应用:行人跟踪

**实验 深度学习与应用:行人跟踪 **


**1、 实验目的**


  • 了解行人跟踪模型基础处理流程

  • 熟悉行人跟踪模型的基本原理

  • 掌握 行人跟踪模型的参数微调训练以及推理的能力

  • 掌握行人跟踪模型对实际问题的应用能力,了解如何在特定的场景和任务中应用该模型

**2、实验环境**


**[镜像详情]**

虚拟机数量:1个(需GPU >=4GB)

虚拟机信息:

  1. 操作系统:Ubuntu20.04

  2. 代码位置:/home/zkpk/experiment/yolo_tracking_main

  3. MOT17数据集存储位置:examples/val_utils/data/MOT17

(数据集下载地址:Https://motchallenge.net)

  1. 已安装软件:python版本:python 3.9,显卡驱动,cuda版本:cuda11.3 cudnn 版本:8.4.1,torch==1.12.1+cu113,torchvision= 0.13.1+cu113

  2. 根据requirements.txt,合理配置python环境

**3、实验内容**


  • 准备多目标跟踪数据集MOT17 ,下载地址位于(Https://motchallenge.net),放置于工程路径为:(examples/val_utils/data/MOT17)

  • 根据不用的行人跟踪算法实现行人跟踪实验

  • 根据实验效果微调行人跟踪算法模型参数

  • 实现离线视频的行人跟踪

**4、实验关键点**


  • 下载数据集放置于指定的文件夹下

  • 配置好算法所需的python虚拟环境

  • 掌握行人跟踪所需的算法基础

  • 具备一定的代码能力,解决实际问题

**5、实验效果图**


行人跟踪效果截图:

![](media/798ashdh.png)

<center>图 1</center>

行人跟踪视频效果:

目标跟踪

**6、实验步骤**


  • 6.1 准备数据集,下载多目标跟踪数据集MOT17 ,下载地址位于(Https://motchallenge.net),将数据集放置于(examples/val_utils/data/MOT17)路径,如下图所示:

<center>图 1</center>

  • 6.2 实现行人跟踪方法对视频的实时检测,运行一下命令进入yolo_tracking_main\examples:

```shell

cd /home/zkpk/experiment/yolo_tracking_main/examples

```

运行python的track.py脚本,命令如下:

```shell

python --yolo-model weights/yolov8n --tracking-method deepocsort ----reid-model weights/lmbn_n_cuhk03_d.pt --source testvideo.mp4 --conf 0.3 --iou 0.5

botsort

strongsort

ocsort

bytetrack

```

分别对应5种不同的目标跟踪模型,实现对行人目标的跟踪

运行日志如下:

```

Successfully loaded imagenet pretrained weights from "weights\osnet_x1_0_imagenet.pth"

video 1/1 (1/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 63.4ms

video 1/1 (2/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (3/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 11.0ms

video 1/1 (4/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (5/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (6/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 11.0ms

video 1/1 (7/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 13.0ms

video 1/1 (8/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 14.0ms

video 1/1 (9/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 14.0ms

video 1/1 (10/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (11/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 13.0ms

video 1/1 (12/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 9.0ms

video 1/1 (13/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 11.0ms

video 1/1 (14/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 14.0ms

video 1/1 (15/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (16/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (17/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 9.0ms

video 1/1 (18/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 13.0ms

video 1/1 (19/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (20/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 14.0ms

video 1/1 (21/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 13.0ms

video 1/1 (22/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 11.0ms

video 1/1 (23/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (24/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 15.0ms

video 1/1 (25/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 11.0ms

video 1/1 (26/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 10.0ms

video 1/1 (27/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 11.0ms

video 1/1 (28/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 13.0ms

video 1/1 (29/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 4 persons, 13.0ms

video 1/1 (30/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 4 persons, 13.0ms

video 1/1 (31/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 14.0ms

video 1/1 (32/2385) E:\PycharmProjects\yolo_tracking_main\examples\testvideo.mp4: 480x640 3 persons, 14.0ms

```

6.3 根据上一步骤6.3 行人跟踪的效果,假如不理想可以使用MOT17数据集微调模型参数(在配置好数据集的情况才可以微调),运行一下命令:

``` shell

python --yolo-model weights/yolov8n.pt --tracking-method deepocsort --benchmark MOT17 --conf 0.45

```

微调参数过程日志如下:

```

2023-11-17 17:34:48.482 | INFO | val:eval:204 - Staring evaluation process on E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1

2023-11-17 17:34:48.560 | INFO | val:eval:204 - Staring evaluation process on E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1

2023-11-17 17:35:00.221 | SUCCESS | boxmot.appearance.reid_model_factory:load_pretrained_weights:207 - Successfully loaded pretrained weights from "E:\PycharmProjects\yolo_tracking_main\examples\weights\osnet_x0_25_msmt17.pt"

2023-11-17 17:35:00.221 | WARNING | boxmot.appearance.reid_model_factory:load_pretrained_weights:211 - The following layers are discarded due to unmatched keys or layer size: ('classifier.weight', 'classifier.bias')

2023-11-17 17:35:00.228 | SUCCESS | boxmot.appearance.reid_model_factory:load_pretrained_weights:207 - Successfully loaded pretrained weights from "E:\PycharmProjects\yolo_tracking_main\examples\weights\osnet_x0_25_msmt17.pt"

2023-11-17 17:35:00.228 | WARNING | boxmot.appearance.reid_model_factory:load_pretrained_weights:211 - The following layers are discarded due to unmatched keys or layer size: ('classifier.weight', 'classifier.bias')

image 1/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000001.jpg: 736x1280 11 persons, 610.4ms

image 1/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000001.jpg: 736x1280 25 persons, 652.3ms

image 2/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000002.jpg: 736x1280 9 persons, 442.2ms

image 2/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000002.jpg: 736x1280 22 persons, 454.5ms

image 3/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000003.jpg: 736x1280 9 persons, 370.0ms

image 3/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000003.jpg: 736x1280 24 persons, 450.9ms

image 4/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000004.jpg: 736x1280 9 persons, 460.8ms

image 4/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000004.jpg: 736x1280 23 persons, 385.0ms

image 5/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000005.jpg: 736x1280 10 persons, 460.4ms

image 5/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000005.jpg: 736x1280 22 persons, 399.6ms

image 6/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000006.jpg: 736x1280 10 persons, 443.0ms

image 7/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000007.jpg: 736x1280 10 persons, 460.8ms

image 6/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000006.jpg: 736x1280 22 persons, 429.7ms

image 8/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000008.jpg: 736x1280 10 persons, 434.5ms

image 7/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000007.jpg: 736x1280 24 persons, 448.3ms

image 9/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000009.jpg: 736x1280 10 persons, 386.9ms

image 8/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000008.jpg: 736x1280 23 persons, 476.7ms

image 10/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000010.jpg: 736x1280 11 persons, 869.1ms

image 9/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000009.jpg: 736x1280 23 persons, 453.9ms

image 11/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000011.jpg: 736x1280 11 persons, 460.8ms

image 10/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000010.jpg: 736x1280 23 persons, 428.2ms

image 12/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000012.jpg: 736x1280 9 persons, 439.9ms

image 11/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000011.jpg: 736x1280 19 persons, 470.3ms

image 13/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000013.jpg: 736x1280 10 persons, 440.8ms

image 12/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000012.jpg: 736x1280 19 persons, 460.8ms

image 14/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000014.jpg: 736x1280 9 persons, 434.2ms

image 13/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000013.jpg: 736x1280 20 persons, 439.0ms

image 15/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000015.jpg: 736x1280 8 persons, 384.9ms

image 14/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000014.jpg: 736x1280 20 persons, 440.8ms

image 16/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000016.jpg: 736x1280 8 persons, 462.8ms

image 15/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000015.jpg: 736x1280 20 persons, 451.8ms

image 17/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000017.jpg: 736x1280 8 persons, 470.7ms

image 16/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000016.jpg: 736x1280 22 persons, 486.0ms

image 18/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000018.jpg: 736x1280 7 persons, 410.9ms

image 17/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000017.jpg: 736x1280 23 persons, 425.9ms

image 19/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000019.jpg: 736x1280 7 persons, 380.0ms

image 20/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000020.jpg: 736x1280 8 persons, 436.8ms

image 18/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000018.jpg: 736x1280 23 persons, 447.8ms

image 21/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000021.jpg: 736x1280 8 persons, 476.0ms

image 19/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000019.jpg: 736x1280 21 persons, 518.6ms

image 22/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000022.jpg: 736x1280 8 persons, 360.0ms

image 20/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000020.jpg: 736x1280 22 persons, 388.6ms

image 23/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000023.jpg: 736x1280 9 persons, 391.0ms

image 21/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000021.jpg: 736x1280 22 persons, 416.9ms

image 24/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000024.jpg: 736x1280 9 persons, 458.8ms

image 22/1050 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-04-FRCNN\img1\000022.jpg: 736x1280 22 persons, 404.7ms

image 25/600 E:\PycharmProjects\yolo_tracking_main\examples\val_utils\data\MOT17\train\MOT17-02-FRCNN\img1\000025.jpg: 736x1280 9 persons, 443.9ms

```

**7、思考题**


  • 考虑在行人跟踪中,模型算法还有哪些改进点

  • 思考怎么将跟踪算法模型应用到手部动作跟踪中

  • 思考如何调节模型参数和训练参数提升模型的效果指标

**8、 实验报告**


请按照实验报告的格式要求撰写实验报告。

相关推荐
HPC_fac130520678161 小时前
以科学计算为切入点:剖析英伟达服务器过热难题
服务器·人工智能·深度学习·机器学习·计算机视觉·数据挖掘·gpu算力
小陈phd3 小时前
OpenCV从入门到精通实战(九)——基于dlib的疲劳监测 ear计算
人工智能·opencv·计算机视觉
Guofu_Liao4 小时前
大语言模型---LoRA简介;LoRA的优势;LoRA训练步骤;总结
人工智能·语言模型·自然语言处理·矩阵·llama
ZHOU_WUYI8 小时前
3.langchain中的prompt模板 (few shot examples in chat models)
人工智能·langchain·prompt
如若1238 小时前
主要用于图像的颜色提取、替换以及区域修改
人工智能·opencv·计算机视觉
老艾的AI世界9 小时前
AI翻唱神器,一键用你喜欢的歌手翻唱他人的曲目(附下载链接)
人工智能·深度学习·神经网络·机器学习·ai·ai翻唱·ai唱歌·ai歌曲
DK221519 小时前
机器学习系列----关联分析
人工智能·机器学习
Robot2519 小时前
Figure 02迎重大升级!!人形机器人独角兽[Figure AI]商业化加速
人工智能·机器人·微信公众平台
浊酒南街10 小时前
Statsmodels之OLS回归
人工智能·数据挖掘·回归
畅联云平台10 小时前
美畅物联丨智能分析,安全管控:视频汇聚平台助力智慧工地建设
人工智能·物联网