Object Tracking

4.

Multiple Object Tracking

4.1.

Problem Formulation

4.1.1.

Definition of Multiple Targets

4.1.2.

Output Requirements

4.1.2.1.

4.1.2.2.

4.2.

Tracking-by-Detection Paradigm

4.2.1.

Object Detection in Each Frame

4.2.1.1.

Detector Selection and Performance

4.2.1.2.

Detection Confidence Scores

4.2.2.

Data Association Across Frames

4.2.2.1.

Cost Matrix Formulation

4.2.2.1.1.

Distance Metrics

4.2.2.1.2.

Feature-based Costs

4.2.2.2.

Association Algorithms

4.2.2.2.1.

Hungarian Algorithm

4.2.2.2.1.1.

Optimal Assignment

4.2.2.2.1.2.

Computational Complexity

4.2.2.2.2.

Greedy Assignment

4.2.2.2.2.1.

Heuristic Approaches

4.2.2.2.2.2.

4.2.2.3.

Association Cues

4.2.2.3.1.

Motion Prediction

4.2.2.3.1.1.

Predicting Future Positions

4.2.2.3.2.

Appearance Similarity

4.2.2.3.2.1.

Feature Extraction for Re-identification

4.2.2.3.3.

Spatial Proximity

4.2.2.3.3.1.

Overlap-based Association

4.3.

Common MOT Frameworks

4.3.1.

Simple Online and Realtime Tracking

4.3.1.1.

Kalman Filter-based Motion Model

4.3.1.2.

IoU-based Association

4.3.2.

4.3.2.1.

Deep Appearance Descriptor

4.3.2.2.

Enhanced Data Association

4.4.

Joint Detection and Tracking

4.4.1.

End-to-End Learning Approaches

4.4.1.1.

Unified Network Architectures

4.4.1.2.

Joint Loss Functions

4.4.2.

Attention and Transformer-based Models

4.4.2.1.

Temporal Attention

4.4.2.2.

Spatial Attention

4.4.2.3.

Sequence-to-sequence Prediction

Previous

3. Single Object Tracking

Go to top

Next

5. Key Challenges in Object Tracking