superdianuj / yolo_detection_tracking Goto Github PK

This work utilizes several open source repos on YOLO based detection and tracking (using DeepSORT), in a crude and lazy manner, to serve as reference. Furthermore, it exposes a fundamental problem with detection and tracking. Detection is significantly better than tracking, and occlusion very easily seeps over time within a video.

Following are the models used in the repos: Models

Following are the resultant video output comparison.

No.	Video Results	Snapshot
1	Reference Video
2	Video Output: occlusion 1_1
3	Video Output: occlusion 1_2
4	Video Output: occlusion 1_3
5	Videeo Output: occlusion 1_4
6	Video Output: occlusion 2
7	Video Output: occlusion 3_1
8	Video Output: occlusion 3_2
9	Video Output: occlusion 4
10	Video Output: occlusion 5
11	Video Output: occlusion 6
12	Video Output: occlusion 7
13	Video Output: occlusion 8_1
14	Video Output: occlusion 8_2
15	Video Output: occlusion 9
16	Video Output: occlusion 10
17	Video Output: occlusion 11