Monocular VO Based on Deep Siamese Convolutional Neural Network

Figure 8

Average errors tested on translation and rotation against different path lengths and speeds. The DSCNN-VO, CNN-VO, and CNN-LSTM-VO were trained and tested in the same conditions. (a) Translation error against path length. (b) Rotation error against path length. (c) Translation error against speed. (d) Rotation error against speed.

(a)

(b)

(c)

(d)