Design of multi-target tracking algorithm based on deep learning

2021 COE Engineering Design Project (XZ01)

---

Faculty Lab Coordinator

Xiao-Ping Zhang

Topic Category

Software Systems

Preamble

In the multi-target tracking problem, the algorithm needs to match the existing target trajectory according to the detection result of the target in each image; for new targets, new targets need to be generated; for targets that have left the camera's field of view, the trajectory needs to be terminated tracking.

Objective

Python language learning Deep learning algorithm learning Target tracking algorithm learning

Partial Specifications

Know the difference between object detection on each image from a video and object tracking, like time costing, and decide which way to use.

Suggested Approach

Learn deep learning models, such as RNN, for tracking.
Learn deep learning models for object detection, such as YOLO.
Design a deep learning based multi-object tracking model, with video as the input, and the multiple bounding boxes as the output.
Learn different features of multi-objects, such as appearance, motion, and interactions.

The paper [1] and related code [2][3] can be reference.
[1] Maksai, Andrii, and Pascal Fua. "Eliminating exposure bias and metric mismatch in multiple object tracking." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019.
[2] https://github.com/maksay/seq-train
[3] https://github.com/abhineet123/Deep-Learning-for-Tracking-and-Detection
[4] Pal, Sankar K., et al. "Deep learning in multi-object detection and tracking: state of the art." Applied Intelligence (2021): 1-30.

Group Responsibilities

Coordinate to define hardware and software interfaces and to divide tasks. Realize and demonstrate the entire system. Deliver the design documents for each module and interfaces.

Student A Responsibilities

Learn deep learning models for multi-object tracking.
Collect data, such as multi-people, or multi-object, collaborate with all the students.
Design a deep learning model for multi-object tracking, with all the students.
Design the model for object detection, to predict Intersection over Union (IoU), with Student B.

Student B Responsibilities

Learn deep learning models for multi-object tracking.
Collect data, such as multi-people, or multi-object, collaborate with all the students.
Design a deep learning model for multi-object tracking, with all the students.
Design the model for object detection, to predict Intersection over Union (IoU), with Student A.

Student C Responsibilities

Learn deep learning models for multi-object tracking.
Collect data, such as multi-people, or multi-object, collaborate with all the students.
Design a deep learning model for multi-object tracking, with all the students.
Design the model for multi-object tracking, by regressing bounding box shift to obtain the ground truth bounding boxes, collaborate with Student D.

Student D Responsibilities

Learn deep learning models for multi-object tracking.
Collect data, such as multi-people, or multi-object, collaborate with all the students.
Design a deep learning model for multi-object tracking, with all the students.
Design the model for multi-object tracking, by regressing bounding box shift to obtain the ground truth bounding boxes, collaborate with Student C.

Course Co-requisites

To ALL EDP Students

Due to COVID-19 pandemic, in the event University is not open for in-class/in-lab activities during the Winter term, your EDP topic specifications, requirements, implementations, and assessment methods will be adjusted by your FLCs at their discretion.