Fig. 1. Schematic diagram of the binocular measurement system based on laser speckle projector

Fig. 2. Convolutional neural network architecture

Fig. 3. Gray distribution pattern of speckle feature points

Fig. 4. Main modules in the backbone. (a) Main downsampling modules; (b) main upsampling modules

Fig. 5. Dynamic depth separable convolution module

Fig. 6. Dynamic convolution block

Fig. 7. Synthetic speckle dataset. (a) Bright condition; (b) dark condition

Fig. 8. Real training dataset. (a) Bright and positive angle-of-view; (b) dark and positive angle-of-view; (c) bright and side angle-of-view; (d) bright and positive angle-of-view with shift change; (e) bright and positive angle-of-view with rotation transformation; (f) bright and positive angle-of-view with angle rotation and scaling transformation

Fig. 9. Experimental system

Fig. 10. Training process. (a) Changes in loss; (b) changes in precision; (c) changes in recall

Fig. 11. Speckle feature point extraction and matching results of the model for each training. (a) Extraction and matching results of pre-trained model in dark condition; (b) extraction and matching results of pre-trained model in bright condition; (c) extraction and matching results of the model after one round of training in dark condition; (d) extraction and matching results of the model after one round of training in bright condition; (e) extraction and matching results of the model after two rounds of training in dark condition; (f) extraction and matching results of the model after two rounds of training in bright condition

Fig. 12. Matching results of each model on real speckle dataset, where the top three rows show dark environmental conditions and the bottom two rows show bright environmental conditions. Each row of helmets has different views

Fig. 13. Measurement experiment of ladder blocks. (a) Collected image by left camera; (b) collected image by right camera; (c) single frame reconstruction of point-cloud of ladder block; (d) reconstruction results of ladder block at different positions

Fig. 14. Measurement experiment of marble plane. (a) Collected image by left camera; (b) collected image by right camera;

Fig. 15. Measurement experiment of helmet. (a) Collected image by left camera; (b) collected image by right camera; (c) reconstruction results

Fig. 16. Measurement experiment of workbench. (a) Handheld measurement system; (b) collected image by left camera; (c) collected image by right camera; (d) reconstruction results

Fig. 17. Comparison between standard convolution and dynamic convolution. (a) Rotation of 90°; (b) rotation of 180°; (c) rotation of 270°