Table 4 FBs processing times on ARM, NEON and DSP processors
From: Real time simultaneous localization and mapping: towards low-cost multiprocessor embedded systems
Functional bloc (FB) | Nonoptimized implementation (μ s) ARM only | Optimized implementation (μs) | |||
|---|---|---|---|---|---|
Processing time per iteration | Mean processing time per frame | Processing time per iteration | Mean processing time per frame | Processing unit | |
Functional bloc (FB) | Processing timeper iteration | Mean processingtime per frame | Processing timeper iteration | Mean processingtime per frame | Processing unit |
1. Prediction | 93 | 93 | 93 | 93 | ARM |
2. FAST | 3400 | 3400 | 3400 | 3400 | ARM |
3. Landmark projection | 9 | 180 | 9 | 180 | ARM |
4. ZMSSD-M | 11.29 | 2630 | 1.27 | 295 | NEON |
5. H i | 14.5 | 66 | 14.5 | 66 | ARM |
6. Estimation | 88845 | 70568 | 15690 | 12552 | NEON |
Initialization task (FB7, 8 and 9) | 3992 | 3922 | 4025 | 4025 | DSP |
Total | – | 80859 | – | 16586 | – |