multi-object-tracker

Introduction to Multi-Object Trackers in Python

The Multi-Object Tracker project offers an easy-to-use implementation of various multi-object tracking algorithms in Python. This tool is essential for those interested in computer vision and requires tracking multiple objects across video frames. By combining different algorithms and methods, the project provides a diverse toolkit for effective and efficient tracking tasks.

Available Multi-Object Trackers

The project features several tracking algorithms, which can be selected based on specific application needs:

• CentroidTracker: A simple yet effective tracking approach that uses the centroid of an object to track it across frames.
• IOUTracker: Utilizes Intersection over Union (IoU) metrics to maintain object tracking.
• CentroidKF_Tracker: Integrates Kalman Filter with CentroidTracker for enhanced prediction and smoothing.
• SORT: A Simple Online and Realtime Tracking algorithm known for its effectiveness in object detection and tracking.

Available OpenCV-based Object Detectors

To facilitate object detection, which is a crucial step before tracking, the project includes several object detectors:

• TF_SSDMobileNetV2: A deep learning model providing fast and accurate object detection.
• Caffe_SSDMobileNet: Another SSD-based model compatible with Caffe framework, known for its efficiency.
• YOLOv3: A popular real-time object detection system used widely for its speed and accuracy.

Installation

To use the multi-object tracking algorithms, OpenCV version 3.4.3 or later is required. It can be installed using pip with the following command:

pip install motrackers

Alternatively, the package can be cloned and installed directly from the project's GitHub repository:

git clone https://github.com/adipandas/multi-object-tracker
cd multi-object-tracker
pip install [-e] .

Note on GPU Use: If intending to use neural network models with a GPU, a CUDA-enabled version of OpenCV must be compiled. Guidance for building OpenCV from source is available here and here.

How to Use: Examples

The interface for each tracker is designed to be straightforward and similar. Here is a basic template of how to use a tracker:

from motrackers import CentroidTracker # or other tracker classes
input_data = ...
detector = ...
tracker = CentroidTracker(...) # or other desired tracker
while True:
    done, image = <read(input_data)>
    if done:
        break
    detection_bboxes, detection_confidences, detection_class_ids = detector.detect(image)
    output_tracks = tracker.update(detection_bboxes, detection_confidences, detection_class_ids)
    for track in output_tracks:
        frame, id, bb_left, bb_top, bb_width, bb_height, confidence, x, y, z = track
        print(track)

For more detailed examples, users can refer to the examples folder in the repository. This section includes hands-on samples that can be cloned and executed.

Pretrained Object Detection Models

To enhance the performance and accuracy of object tracking, pretrained weights for neural-network models are available. Users can download these weights using shell scripts provided in the repository. Detailed instructions are contained in DOWNLOAD_WEIGHTS.md.

Notes

• There may be some differences in implementation compared to the original papers for SORT and IoU Tracker.
• Users encountering bugs or issues are encouraged to contribute through pull requests or by opening an issue on the project's GitHub page.

For further information on references and contributions, users can view REFERENCES.md and CONTRIBUTING.md.

Citing the Project

When using this project in any academic or professional work, please consider citing it as follows:

@misc{multiobjtracker_amd2018,
  author = {Deshpande, Aditya M.},
  title = {Multi-object trackers in Python},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/adipandas/multi-object-tracker}},
}

The Multi-Object Tracker project is a powerful resource for anyone looking to implement efficient and accurate object tracking systems. Whether for academic research, industrial application, or personal projects, it offers a flexible framework to meet varied tracking needs.