Benchmark:

MOT20 | MOT17 |

Short name:

CT_v0

Detector:

Public

Description:

ClusterTracker uses DBSCAN to perform spatial clustering on the object detections from short 10-frame sequences, treats clusters as tracklets, and then merges successive tracklets with high bounding box overlap to form tracks. CT can be run in an online mode with a 10 frame latency or in parallel across multiple cores for offline analysis.

Reference:

J. Lohn-Jaramillo, L. Ray, R. Granger, E. Bowen. ClusterTracker: An Efficiency-Focused Multiple Object Tracking Method. In , 2022.

Last submitted:

February 12, 2022 (2 years ago)

Published:

January 30, 2022 at 15:59:23 CET

Submissions:

3

Project page / code:

https://github.com/JLJ19/ClusterTracker

Open source:

No

Hardware:

i7 6700 3.4GHZ, 1 Core

Runtime:

317.9 Hz

Benchmark performance:

Detailed performance:

Raw data:

download