Author Affiliations
1 School of Microelectronics, Tianjin University, Tianjin 300072, China2 Tianjin Key Laboratory of Imaging and Sensing Microelectronic Technology, Tianjin 300072, Chinashow less
Fig. 1. Flow chart of the proposed algorithm
Fig. 2. (a) Background image; (b) scene image; (c) result with fixed threshold; (d) result with adaptive threshold
Fig. 3. Eight experimental scenarios
Fig. 4. Segmentation results obtained by different algorithms for the scene No. 5. (a) Algorithm in Ref. [5]; (b) FCN algorithm; (c) proposed algorithm with preprocessing; (d) proposed algorithm without preprocessing
Fig. 5. Segmentation results obtained by different algorithms for the scene No. 8. (a) Algorithm in Ref. [5]; (b) FCN algorithm; (c) proposed algorithm with preprocessing; (d) proposed algorithm without preprocessing
Parameter | Adaptivethreshold | Fixed threshold |
---|
θS=150 | θS=200 | θS=250 | |
---|
Number of feature points | 50 | 40 | 55 | 89 | Number of mismatching feature points | 1 | 2 | 3 | 13 | Mismatching rate /% | 2.0 | 5.0 | 5.5 | 14.6 |
|
Table 1. Comparison of matching points between adaptive threshold and fixed threshold
Scene | Mismatching rate /% |
---|
Original SIFTalgorithm | Proposed SIFTalgorithm |
---|
1 | 18.9 | 1.5 | 2 | 10.0 | 0 | 3 | 9.3 | 1.7 | 4 | 9.7 | 0 | 5 | 14.3 | 2.0 | 6 | 10.4 | 1.9 | 7 | 22.5 | 1.2 | 8 | 19.8 | 2.7 | Average | 13.2 | 1.4 |
|
Table 2. Comparison of mismatching rate between the proposed SIFT algorithm and the original SIFT algorithm
Scene | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | Average |
---|
Proposedalgorithm (withpreprocessing) | RO /% | 3.40 | 4.38 | 7.49 | 6.47 | 9.84 | 9.86 | 9.18 | 9.28 | 7.49 | RU /% | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | RE /% | 3.40 | 4.38 | 7.49 | 6.47 | 9.84 | 9.86 | 9.18 | 9.28 | 7.49 | Proposedalgorithm(withoutpreprocessing) | RO /% | 3.40 | 4.31 | 7.49 | 6.17 | 8.95 | 9.83 | 7.87 | 8.09 | 7.01 | RU /% | 0 | 1.70 | 0 | 0.85 | 1.29 | 0.26 | 2.65 | 1.20 | 0.99 | RE /% | 3.40 | 6.11 | 7.49 | 7.08 | 10.37 | 10.12 | 10.80 | 9.40 | 8.10 | FCNalgorithm | RO /% | 2.42 | 1.17 | 3.53 | 4.04 | 5.94 | 5.72 | 7.70 | 4.70 | 4.40 | RU /% | 1.12 | 4.36 | 2.95 | 2.08 | 3.34 | 2.54 | 1.82 | 3.81 | 2.74 | RE /% | 3.58 | 6.79 | 6.68 | 6.25 | 9.48 | 8.47 | 9.69 | 8.85 | 7.47 | Algorithmin Ref. [5] | RO /% | 18.19 | 25.00 | 20.39 | 23.26 | 14.26 | 31.88 | 14.11 | 11.77 | 19.86 | RU /% | 0 | 23.12 | 15.79 | 13.04 | 10.20 | 26.67 | 3.23 | 15.44 | 13.44 | RE /% | 18.19 | 62.59 | 42.96 | 41.74 | 27.25 | 79.83 | 17.92 | 32.18 | 40.33 |
|
Table 3. Comparison of segmentation errors among FCN algorithm, the algorithm in Ref. [5] and the proposed algorithms
Algorithm | Preprocessing | Stereomatching | Region refinement basedon depth information | SIFTmatching | Mean shift | Total |
---|
Withpreprocessing | 0.38 | 0.06 | 0.06 | 0.49 | 0.02 | 1.01 | Withoutpreprocessing | | 0.06 | 0.06 | 1.51 | 0.02 | 1.65 |
|
Table 4. Runtime of each stage of the proposed algorithm (unit: s)