• Laser & Optoelectronics Progress
  • Vol. 58, Issue 4, 0415001 (2021)
Zaiteng Zhang, Rongfen Zhang, and Yuhong Liu*
Author Affiliations
  • College of Big Data and Information Engineering, Guizhou University, Guiyang, Guizhou 550025, China
  • show less
    DOI: 10.3788/LOP202158.0415001 Cite this Article Set citation alerts
    Zaiteng Zhang, Rongfen Zhang, Yuhong Liu. Visual Odometry Algorithm Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0415001 Copy Citation Text show less
    Extraction framework of sparse feature method
    Fig. 1. Extraction framework of sparse feature method
    Framework of proposed network
    Fig. 2. Framework of proposed network
    Proposed neural network structure
    Fig. 3. Proposed neural network structure
    Structure of attention module
    Fig. 4. Structure of attention module
    Mean translation errors and mean rotation errors under different conditions. (a) Mean translation errors under different path lengths; (b) mean rotation errors under different path lengths; (c) mean translation errors under different speeds; (d) mean rotation errors under different speeds
    Fig. 5. Mean translation errors and mean rotation errors under different conditions. (a) Mean translation errors under different path lengths; (b) mean rotation errors under different path lengths; (c) mean translation errors under different speeds; (d) mean rotation errors under different speeds
    VO trajectories in different scenarios. (a) Scenario 03; (b) scenario 04; (c) scenario 09
    Fig. 6. VO trajectories in different scenarios. (a) Scenario 03; (b) scenario 04; (c) scenario 09
    Practical experiment. (a) VO trajectory in actual scene; (b) practical map
    Fig. 7. Practical experiment. (a) VO trajectory in actual scene; (b) practical map
    LayerReceptive field sizePaddingStrideNumber of channels
    Conv17×73264
    Conv25×522128
    Conv35×522256
    Conv3_13×311256
    Conv43×312512
    Conv4_13×311512
    Conv53×312512
    Conv5_13×311512
    Conv63×3121024
    Table 1. Parameters of CNN
    AlgorithmConstituteFLOPs /1010Parameter /108trel/%rrel /(°)
    VISO2_MConventional (monocular)\\17.4816.52
    VISO2_SConventional (stereo)\\1.891.96
    DeepVOCNN+LSTM5.143(not include LSTM)5.10165.966.12
    Proposed algorithmCNN+attention5.1862.60318.8515.68
    Table 2. Parameter calculation of different algorithms
    Zaiteng Zhang, Rongfen Zhang, Yuhong Liu. Visual Odometry Algorithm Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0415001
    Download Citation