Kuluum/sort-cpp

C++ version of SORT: Simple online and realtime tracking of multiple objects in a video sequence

Introduction

C++ implementation of SORT: Simple, online, and real-time tracking of multiple objects in a video sequence.

Kuhn-Munkres (Hungarian) Algorithm in C++ is forked from: https://github.com/saebyn/munkres-cpp

Dependencies

• Ubuntu 16.04
• Docker 18.09.4
• OpenCV 3.4.2
• Boost 1.58.0

All of the 3rd party libraries are included in the provided docker image

Build Docker Image

1. Open Dockerfile, change line #19 ARG USERNAME to your host user name (echo $USER)
2. Open a terminal and run:

   $ cd /path/to/sort-cpp
   $ docker build -t sort .
   $ ./docker_run.sh

Demo:

To run the tracker with the provided detections and visualize the results:

1. Download the 2D MOT 2015 benchmark dataset
2. Create a symbolic link to the dataset

   $ ln -s /path/to/MOT2015_challenge/data/2DMOT2015 /path/to/sort-cpp/mot_benchmark

3. Run the demo

   $ cd /path/to/sort-cpp
   $ mkdir build && cd "$_"
   $ cmake .. && make
   $ cd /path/to/sort-cpp/bin
   # Without display
   $ ./sort-cpp
   # With display
   $ ./sort-cpp -d

Evaluate Metrics

Using the Python implementation of metrics for benchmarking multiple object trackers (MOT) to evaluate metrics.

Dataset Structure

Layout for ground truth data
    <GT_ROOT>/<SEQUENCE_1>/gt/gt.txt
    <GT_ROOT>/<SEQUENCE_2>/gt/gt.txt
    ...

Layout for test data
    <TEST_ROOT>/<SEQUENCE_1>.txt
    <TEST_ROOT>/<SEQUENCE_2>.txt
    ...

Example:
mot_benchmark
├── test
│   ├── ADL-Rundle-6.txt
│   └── ADL-Rundle-8.txt
└── train
    ├── ADL-Rundle-6
    │   └── gt
    │       └── gt.txt
    └── ADL-Rundle-8
        └── gt
            └── gt.txt


Sequences of ground truth and test will be matched according to the `<SEQUENCE_X>`
string.

Example

# Optional for virtualenv
$ source ~/env/bin/activate
$ pip install motmetrics
# Usage
$ python -m motmetrics.apps.eval_motchallenge --help
# Format: python -m motmetrics.apps.eval_motchallenge groundtruths tests
$ python -m motmetrics.apps.eval_motchallenge mot_benchmark/train output/

Result

FPS is around 1900 with Debug build.

References

1. https://github.com/abewley/sort
2. https://github.com/mcximing/sort-cpp
3. https://github.com/saebyn/munkres-cpp