Fig. 1. Principle of multi-stage small object detection

Fig. 2. SRCNN network framework

Fig. 3. Comparison of traditional algorithm and optimization algorithm framework. (a) Mainstream multi-scale target recognition architecture; (b) feature aggregation and single scale recognition improvement strategies

Fig. 4. RRT and RRT* search process. (a) RRT; (b) RRT*

Fig. 5. RRT*-Connect with target bias strategy

Fig. 6. Path planning. (a) Path optimization; (b) path smoothing

Fig. 7. Robotic arm and obstacle bounding box model

Fig. 8. Simulated lesion point cloud image. (a) Processed point cloud image; (b) normal vector image of the lesion surface

Fig. 9. Surgical robotic arm guidance system

Fig. 10. System flow chart

Fig. 11. Comparison of recognition results. (a) Intel RealSense Viewer; (b) bright condition; (c) dark condition

Fig. 12. Search process and results. (a) RRT*; (b) RRT-Connect; (c) BT-RRT*; (d) proposed algorithm

Fig. 13. Comparison of simulation results. (a) No constraint strategy applied; (b) constraint strategy applied

Fig. 14. Experimental results in actual scene. (a) No constraint strategy applied; (b) constraint strategy applied