Table 2 Summary of vision-based stair recognition systems for robotic leg prostheses and exoskeletons
From: StairNet: visual recognition of stairs for human–robot locomotion
Reference | Camera | Position | Data set Size | Classifier | Computing Device | Test Accuracy |
|---|---|---|---|---|---|---|
[11] | RGB | Waist | 7284 | Convolutional neural network | NVIDIA Titan X | 99.6% |
[10] | Depth | Chest | 170 | Heuristic thresholding and edge detector | Intel Core i5 | 98.8% |
[9] | Depth | Leg | 8455 | Support vector machine | Intel Core i7-2640M | 98.5% |
StairNet | RGB | Chest | 515,452 | Convolutional neural network | Google Cloud TPU | 98.4% |
[17] | Depth | Leg | 3000 | Convolutional neural network | NVIDIA Quadro P400 | 96.8% |
[8] | Depth | Leg | 109,699 | Cubic kernel support vector machine | Intel Core i7-2640M | 95.6% |
[14] | RGB | Chest | 34,254 | Convolutional neural network | NVIDIA TITAN Xp | 94.9% |
[15] | RGB | Head | 123,979 | Bayesian deep neural network | NVIDIA Jetson TX2 | 93.2% |
[16] | RGB | Leg | 123,954 | Bayesian deep neural network | NVIDIA Jetson TX2 | 92.4% |
(27) | RGB | Chest | 542,868 | Convolutional neural network | Google Cloud TPU | 70.8% |