Author Affiliations
1 School of Electrical Automation and Information Engineering, Tianjin University, Tianjin 300072, China2 School of Microelectronics, Tianjin University, Tianjin 300072, Chinashow less
Fig. 1. Architecture of region proposal network
Fig. 2. Multi-view samples. (a) Frontal view; (b) profile view; (c) back view
Fig. 3. Architecture of MVNN
Fig. 4. Principal detection neural network model of MVNN
Fig. 5. Three channels of input layer. (a) Sample 1; (b) sample 2
Fig. 6. Calculation model of part score for deformation layer
Fig. 7. Comparison results of region proposal for input data. (a) HOG+Adaboost algorithm; (b) Proposed region proposal algorithm
Fig. 8. Testing result of multi-view model. (a) Testing result of multiple views; (b) Comparison results of single-view model and multi-view model
Fig. 9. Testing result of DPM
Fig. 10. Testing result of proposed algorithm on IHDD. (a) RFPPI-RMR curve; (b) P-R curve
Parameter | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 |
---|
Starting line | 1 | 1 | 4 | 4 | 1 | 1 | 1 | 4 | 1 | 1 | 1 | 1 | -1 | 1 | 1 | Ending line | 3 | 3 | 9 | 9 | 3 | 9 | 9 | 9 | 3 | 9 | 9 | 9 | 9 | 9 | 9 | Starting column | 1 | 3 | 1 | 3 | 1 | 1 | 4 | 1 | 1 | 1 | 1 | 4 | 1 | 1 | 1 | Ending column | 3 | 5 | 3 | 5 | 5 | 2 | 5 | 5 | 5 | 5 | 2 | 5 | 5 | 5 | 5 |
|
Table 1. Parameters of part filters in proposed algorithm
Test environment | Total frames | Annotated humansNo. |
---|
Office day 1 | 3912 | 5163 | Office day 2 | 3157 | 4785 | Office day 3 | 246 | 394 | Duty room day 1 | 108 | 178 | Duty room day 2 | 813 | 893 | Duty room night 1 | 6063 | 6072 | Duty room night 2 | 369 | 369 | Training set | 8799 | 10479 | Validation set | 2933 | 3701 | Test set | 2933 | 3674 | Total number of samples | 14665 | 17854 |
|
Table 2. Indoor human detection dataset
Parameter | Epoch 1-150 | Epoch 151-250 |
---|
Learning rate | 0.025 | 0.0125 | Momentum | 0.9 | 0.9 | Batch size | 80 | 80 |
|
Table 3. Parameter setting of second-stage network model
Algorithm | RMR | RAP /% |
---|
measurements | (RFPPI-RMR) /% | | Ref. [3] | 37.38 | 57.32 | Ref. [21] | 17.29 | 84.82 | Ref. [18] | 28.84 | 75.52 | Proposed | 14.66 | 87.34 |
|
Table 4. Quantitative comparison of different algorithms on dataset