Now You See Them, Now You Dont FusionGhost Gait for Occlusion-Robust Gait Recognition with ArabGait dataset
Organizations: Department of Computer Science and Engineering, American University of Sharjah, Sharjah, United Arab Emirates
Abstract
Skeleton-based gait recognition is generally more robust to appearance changes than pixel-based methods, but this robustness depends on stable pose extraction. Clothing variation in gait recognition includes changes in garments, silhouette, and carried items; under Gulf clothing (Kandoora/Abaya), the dominant failure mode shifts from appearance change to lower-limb pose degradation caused by severe occlusion. To study this regime, we introduce ArabGait, a 45-subject indoor multi-view dataset with four walking conditions across three viewpoints, and show that Gulf clothing lowers mean lower-limb confidence from 0.811 to 0.683 and raises lower-limb failure rate from 7.98% to 17.82%, with ankles most affected. We propose FusionGhost-Gait, a fused spatiotemporal architecture that integrates spatial graph blocks and temporal modeling in one residual unit, removes redundant channel projections, and uses compact temporal Ghost kernels as a lightweight regularization mechanism for pose-noisy trajectories while preserving efficient temporal downsampling. On ArabGait cross-clothing evaluation, including Protocol A and Protocol B, FusionGhost-Gait reaches up to 67.91% Rank-1 while remaining compact (108,970- 133,674 parameters and 0.035-0.052 billion floating-point operations (GFLOPs)), achieving more than 65% fewer parameters and 75% lower FLOPs than existing baselines. These results indicate that the proposed model generalizes effectively to challenging Gulf-clothing conditions while remaining lightweight and deployable for surveillance and access-control biometric systems.
Figures & tables
| Property | Value |
|---|---|
| Number of subjects | 45 |
| Age range | 18–25 years |
| Gender distribution | 27 men and 18 women |
| Walking conditions | 4 (Casual clothing, Casual clothing with Bag, |
| Gulf clothing, Gulf clothing with Bag) | |
| Camera views | 3 (45°, 90°, 135°) |
| Condition | Upper Conf. | Lower Conf. | Lower Fail (%) | Lower Std. |
|---|---|---|---|---|
| Casual clothing | 0.831 | 0.811 | 7.98 | 0.046 |
| Casual clothing with Bag | 0.801 | 0.766 | 11.83 | 0.056 |
| Gulf clothing | 0.780 | 0.729 | 13.03 | 0.059 |
| Gulf clothing with Bag | 0.740 | 0.683 | 17.82 | 0.064 |
| Hyperparameter | Value |
|---|---|
| Optimizer | AdamW |
| Learning rate | 0.005 |
| Weight decay | |
| Training loss | Supervised Contrastive ( Khosla and others, 2020 ) |
| Temperature ( ) | 0.01 |
| Distance metric | (Euclidean) |
| Type | Method | NM (%) | BG (%) | CL (%) |
|---|---|---|---|---|
| Appearance-based | GaitNet ( Song et al., 2019 ) | 91.6 | 85.7 | 58.9 |
| GaitSet ( Chao et al., 2019 ) | 95.0 | 87.2 | 70.4 | |
| GaitPart ( Fan et al., 2020 ) | 96.2 | 91.5 | 78.7 | |
| Model-based | PoseGait ( Liao et al., 2020b ) | 68.7 | 44.5 | 36.0 |
| GaitGraph2 ( Teepe et al., 2022 ) | 87.7 | 74.8 | 66.3 | |
| GaitGraph ( Teepe et al., 2021 ) | 82.0 | 73.2 | 63.6 |
| Method | Params | GFLOPs |
|---|---|---|
| GaitNet ( Song et al., 2019 ) | 16.19M | 39.78 |
| PoseGait ( Liao et al., 2020b ) | 4.229M | 0.096 |
| GaitSet ( Chao et al., 2019 ) | 2.59M | 8.6 |
| GaitPart ( Fan et al., 2020 ) | 1.204M | 6.403 |
| GaitGraph ( Teepe et al., 2021 ) | 425,834 | 0.233 |
| GaitGraph2 ( Teepe et al., 2022 ) | 350,822 | 0.206 |
| Model | Protocol A | Protocol B | ||||
|---|---|---|---|---|---|---|
| Arch | Overall | G | GB | Overall | C | CB |
| ResGCN-N21 | 59.56 | 61.11 | 58.01 | 65.22 | 58.12 | 72.31 |
| FusionGhost-Gait-Nano | 71.51 | 70.59 | 72.44 | 70.75 | 65.94 | 75.57 |
| ResGCN-N39 | 62.81 | 65.36 | 60.26 | 68.19 | 63.75 | 72.64 |
| ResGCN-N51 | 56.33 | 58.82 | 53.85 | 59.44 | 54.38 | 64.50 |
| ResGCN-N57 | 57.15 | 59.80 | 54.49 | 56.27 | 50.00 | 62.54 |
| Model | Protocol A | Protocol B | ||||
|---|---|---|---|---|---|---|
| Arch | Overall | G | GB | Overall | C | CB |
| ResGCN-N21 | 62.17 | 65.69 | 58.65 | 67.07 | 63.12 | 71.01 |
| FusionGhost-Gait-Nano | 64.92 | 68.30 | 61.54 | 65.00 | 60.94 | 69.06 |
| ResGCN-N39 | 52.28 | 53.59 | 50.96 | 49.52 | 45.62 | 53.42 |
| ResGCN-N51 | 39.83 | 41.83 | 37.82 | 38.63 | 37.19 | 40.07 |
| ResGCN-N57 | 52.63 | 56.86 | 48.40 | 54.30 | 50.62 | 57.98 |
| Stream | Casual Clothing | Casual Clothing with Bag | Gulf Clothing | Gulf Clothing with Bag | Gulf vs Casual (%) |
|---|---|---|---|---|---|
| Joints | 132.27 | 139.15 | 146.55 | 154.66 | % |
| Bones | 9.89 | 11.28 | 13.21 | 14.50 | % |
| Velocity | 108.54 | 120.65 | 131.39 | 147.57 | % |
Gait Recognition
Gait recognition identifies people from their walking patterns, using visual appearance and motion cues across cameras and other sensors.
See all 32 papers