Benchmarking Deep Learning Models for Wheelchair Configuration Classification in Accessible Automated Vehicle Applications
Organizations: University of California Merced
Abstract
Making automated vehicles accessible requires consideration of vulnerable road users, including wheelchair users. To offer a more tailored experience that supports independence for this group, vehicle design needs to accommodate the wide range of wheelchair configurations. However, wheelchairs are individualized, varying in frame geometry, propulsion, and visual appearance. Moreover, occupants or accessories may obscure visual cues. Existing mobility-aid datasets only label the presence of a wheelchair, rather than describing its specific configuration. To fill this gap, this paper introduces the Wheelchair Configuration Image Dataset (WCID), comprising 467 face-blurred images across 192 capture groups solely for evaluation. Each image is annotated with (i) a two-class label of manual or powered, (ii) a chair subtype, (iii) occupancy, viewpoint, setting, lighting, occlusion, propulsion, and control, with a 273-image validation split and 194-image test split. Three convolutional neural network backbones and a frozen vision-language probe are trained on an external image dataset and evaluated on WCID to measure cross-dataset generalization, while also testing robustness using Gaussian blur. Convolutional models do well on clean images, but performance drops under strong blur, leaving a gap before deployment. Code, split assignments, labels, trained weights, and the face-blurred images is available at: https://github.com/Mi3-Lab/MAID
Figures & tables
| Class (prediction target) | |
|---|---|
| Manual | Wheelchairs with no integrated electric propulsion; self-propelled via hand rims on large rear wheels, or attendant-propelled. |
| Powered | Wheelchairs with electric propulsion; visible drive wheels, battery housing, joystick or similar control. |
| Subtype (descriptive) | |
| Manual | |
| Standard | Conventional frame. |
| Negative camber | Rear wheels angled inward at the top. |
| Field | Cohen’s | Field | Cohen’s | ||
|---|---|---|---|---|---|
| Occupancy | 1.00 | 102 | Setting | 0.59 | 102 |
| Device type | 0.86 | 102 | Viewpoint | 0.45 | 102 |
| Propulsion | 0.76 | 102 | Occlusion | 0.05 | 102 |
| Lighting | 0.73 | 102 |
| Split | Img. | Groups | Class counts | Image source |
|---|---|---|---|---|
| Val | 273 | 83 | manual 207, powered 66 | OI 86, OV 187 |
| Test | 194 | 109 | manual 147, powered 47 | OI 176, OV 18 |
| All | 467 | 192 | manual 354, powered 113 | OI 262, OV 205 |
| Model | Acc. | Acc. 95% CI | Bal. acc. | Macro-F1 | F1 95% CI | Manual Recall | Powered Recall |
|---|---|---|---|---|---|---|---|
| ResNet-50 | [0.818, 0.929] | [0.761, 0.901] | |||||
| EfficientNet-B0 | [0.774, 0.892] | [0.697, 0.846] | |||||
| ConvNeXt-Tiny | [0.883, 0.956] | [0.833, 0.937] | |||||
| SigLIP 2 (frozen) | [0.681, 0.823] | [0.605, 0.770] |
| Model | Blur | |||||
|---|---|---|---|---|---|---|
| ResNet-50 | Before | |||||
| ResNet-50 | After | |||||
| EfficientNet-B0 | Before | |||||
| EfficientNet-B0 | After | |||||
| ConvNeXt-Tiny | Before | |||||
| ConvNeXt-Tiny | After |
| Model | AURC | Acc.@90% | Acc.@80% | Acc.@70% |
|---|---|---|---|---|
| ResNet-50 | ||||
| EfficientNet-B0 | ||||
| ConvNeXt-Tiny | ||||
| SigLIP 2 (frozen) |
Classification
Classification maps observations to discrete labels, including familiar, unknown, hierarchical, or multiple categories.
See all 451 papers