DeepSORT

lastmod 2024-02-20created 2019-03-125 edits

SORT (Simple Online and Realtime Tracking) is a simple and effective algorithm for performing MOT (Multiple Object Tracking). DeepSORT improves on SORT's performance by integrating appearance information (Appearance Information) into the existing approach. This new approach makes it possible to track objects even through long periods of occlusion, which effectively reduces how often ID switches occur. To carry out the new technique proposed by DeepSORT, a Deep Associatoin Metric is trained on a Large-scale Person Re-identification Dataset in the pre-training stage.

In short, DeepSort is an object tracking framework that tracks object IDs more accurately.

deep_sort_realtime example

from deep_sort_realtime.deepsort_tracker import DeepSort

tracker = DeepSort(max_age=50)

Import the DeepSort-related library. DeepSort is set as the tracker, and max_age means that an object is no longer tracked if it has not been detected for longer than max_age.

Projects

• Simple Online and Realtime Tracking with a Deep Association Metric

See also

• Motion estimation
• Simple Online and Realtime Tracking (SORT)
• Deep learning

Favorite site

• Github - Deep SORT
• Reddit - DeepSORT: Deep Learning to Track Custom Objects in a Video
• DeepSORT: Deep Learning to Track Custom Objects in a Video
• DeepSORT - Papers
• Building a person tracker with DeepSORT
• DeepSORT: Deep Learning to Track Custom Objects in a Video
• Deep Sort – JM LOG
• [Recommended] Running YOLOv8 with OpenCV (YOLOv8 + DeepSort)
• DeepSORT: Simple Online and Realtime Tracking with a Deep Association Metric paper review | by Heejun Park | Medium
• Understanding DeepSORT properly