Types

Approach 1

1. • Single Object tracking (SOT) 

   • Multiple Object Tracking (MOT)

Approach 2

1. • Online Tracking: using several frames in a video

   • Offline Tracking: using all frames of a video. <= more accurate.

Approach 3

1. • Detection-based Tracking: object detection + tracking.

   • Detection-free Tracking: Apply RNN, LSTM.

Traditional Methods

1. Meanshift (Mode seeking)

2. Optical flow

Deep Learning-based Approaches

1. Deep Regression Networks (ECCV, 2016)

2. ROLO - Recurrent Yolo (ISCAS 2016)

3. SORT - Deep SORT

Targets

1. Bounding Box

2. Object IDs

3. ID maintenance when the object is lost in a short time (several frames)

4. Real-time

5. Applicability