DSFC: Deep Sort Optical Flow Compensation with YOLOv4
TUD-Crossing
Short name:
DSFC
Detector:
Private
Description:
we design a motion flow and compensation strategy for predicting the location of occluded objects. Since the occluded objects may be legible in earlier frames, we utilize the velocity and location of the objects in the past frames to predict the possible location of the occluded objects. In addition, to improve the tracking speed and further enhance the tracking robustness, we utilize efficient YOLOv4, train on humanCrowd dateset to produce the detections in the proposed algorithm. By using YOLOv4, the tracking speed of our proposed method improved significantly.
Reference:
Last submitted:
November 16, 2021 (2 months ago)
Published:
November 19, 2021 at 11:52:07 CET
Submissions:
1
Project page / code:
n/a
Open source:
No
Hardware:
2.4 GHZ, 1 Core
Runtime:
10.8 Hz
Benchmark performance:
| Sequence | MOTA | IDF1 | HOTA | MT | ML | FP | FN | Rcll | Prcn | AssA | DetA | AssRe | AssPr | DetRe | DetPr | LocA | FAF | ID Sw. | Frag |
| MOT15 | 55.1 | 59.5 | 45.7 | 228 (31.6) | 166 (23.0) | 5,786 | 21,301 | 65.3 | 87.4 | 45.1 | 46.7 | 49.9 | 73.0 | 52.4 | 70.2 | 80.5 | 1.0 | 504 (0.0) | 1,671 (0.0) |
Detailed performance:
| Sequence | MOTA | IDF1 | HOTA | MT | ML | FP | FN | Rcll | Prcn | AssA | DetA | AssRe | AssPr | DetRe | DetPr | LocA | FAF | ID Sw. | Frag |
| ADL-Rundle-1 | 43.5 | 56.1 | 43.2 | 11 | 3 | 1,809 | 3,400 | 63.5 | 76.6 | 45.9 | 41.1 | 50.2 | 73.9 | 50.1 | 60.5 | 79.5 | 3.6 | 49 | 440 |
| ADL-Rundle-3 | 47.0 | 53.0 | 42.5 | 10 | 8 | 698 | 4,633 | 54.4 | 88.8 | 42.5 | 42.5 | 45.0 | 80.1 | 46.3 | 75.6 | 84.5 | 1.1 | 57 | 147 |
| AVG-TownCentre | 60.3 | 72.8 | 49.9 | 86 | 27 | 636 | 2,125 | 70.3 | 88.8 | 52.6 | 47.8 | 58.2 | 71.3 | 53.0 | 67.0 | 76.5 | 1.4 | 76 | 352 |
| ETH-Crossing | 69.1 | 75.1 | 56.2 | 4 | 6 | 17 | 286 | 71.5 | 97.7 | 55.8 | 56.7 | 58.7 | 81.2 | 59.6 | 81.4 | 83.7 | 0.1 | 7 | 32 |
| ETH-Jelmoli | 60.4 | 68.7 | 54.4 | 19 | 11 | 347 | 640 | 74.8 | 84.5 | 54.7 | 54.3 | 61.3 | 78.2 | 63.4 | 71.7 | 84.4 | 0.8 | 17 | 40 |
| ETH-Linthescher | 53.4 | 56.9 | 46.0 | 42 | 95 | 296 | 3,816 | 57.3 | 94.5 | 46.3 | 45.6 | 51.3 | 77.8 | 48.3 | 79.8 | 84.2 | 0.2 | 52 | 84 |
| KITTI-16 | 58.6 | 72.3 | 50.6 | 7 | 1 | 164 | 525 | 69.1 | 87.8 | 53.2 | 48.3 | 57.2 | 70.3 | 53.5 | 67.9 | 76.8 | 0.8 | 15 | 29 |
| KITTI-19 | 53.2 | 63.8 | 45.6 | 16 | 10 | 740 | 1,728 | 67.7 | 83.0 | 46.4 | 45.1 | 51.5 | 70.1 | 51.6 | 63.3 | 75.6 | 0.7 | 35 | 133 |
| PETS09-S2L2 | 64.8 | 48.7 | 38.5 | 18 | 1 | 973 | 2,235 | 76.8 | 88.4 | 27.7 | 53.6 | 32.8 | 56.6 | 60.5 | 69.6 | 78.9 | 2.2 | 183 | 286 |
| TUD-Crossing | 82.8 | 80.5 | 57.9 | 9 | 0 | 4 | 180 | 83.7 | 99.6 | 54.2 | 62.0 | 61.5 | 71.9 | 64.9 | 77.2 | 79.7 | 0.0 | 6 | 37 |
| Venice-1 | 59.6 | 64.2 | 50.0 | 6 | 4 | 102 | 1,733 | 62.0 | 96.5 | 51.9 | 48.3 | 58.1 | 75.2 | 51.0 | 79.3 | 83.1 | 0.2 | 7 | 91 |
Raw data:
