Apple-MOT

---

Code of paper: Research on UAV Online Visual Tracking Algorithm based on YOLOv5 and FlowNet2 for Apple Yield Inspection

---

Abstract

• A real-time apple tracking and yield estimation algorithm
• YOLOv5 and FlowNet2 are integrated into Tracking-by-Detecting framework
• The accuracy of apple detection is 85.5%
• Apple tracking and counting accuracy is 92.51%

Quick Start

Python requirements

• python == 3.10.4
• numpy == 1.22.3
• pytorch == 1.11.0
• torchvision == 0.12.0
• cuda == 11.3.1
• cudnn == 8.2.1
• opencv == 4.5.4
• tensorboard == 2.8.0

Installation

Download the code

git clone https://github.com/wangwang-xyz/Apple-MOT.git

Download the data in here. Extract code: fisi

Add your GPU into FlowNet2 config files:

• flownet2/networks/channelnorm_package/setup.py
• flownet2/networks/correlation_package/setup.py
• flownet2/networks/resample2d_package/setup.py

nvcc_args = [
    '-gencode', 'arch=compute_50,code=sm_50',
    '-gencode', 'arch=compute_52,code=sm_52',
    '-gencode', 'arch=compute_60,code=sm_60',
    '-gencode', 'arch=compute_61,code=sm_61',
    '-gencode', 'arch=compute_70,code=sm_70',
    '-gencode', 'arch=compute_86,code=sm_86',
    '-gencode', 'arch=compute_70,code=compute_70'
    # '-gencode', 'arch=compute_XX,code=sm_XX',
    # you can check in Nvidia website
]

Then install flownet2

cd flownet2
bash install.sh

At last, install yolov5

cd ..
cd yolov5
pip install -r requirements.txt  # install

The original pre-trained parameters for YOLOv5 and FlowNet2 can be downloaded from their github websites, respectively.

Inference

The core code is written in src/

Change the video root in track.py before run it

cd ..
cd src
python tracker.py