Author Affiliations
1 Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang, Liaoning 110016, China2 Institutes for Robotics and Intelligent Manufacturing, Chinese Academy of Sciences, Shenyang, Liaoning 110016, China3 University of Chinese Academy of Sciences, Beijing 100049, China4 Key Laboratory of Opto-Electronic Information Processing, Chinese Academy of Sciences, Shenyang, Liaoning 110016, China5 Liaoning Provincial Key Laboratory of Image Understanding and Computer Vision, Shenyang, Liaoning 110016, Chinashow less
Fig. 1. Overview of improved system
Fig. 2. Schematic of four-layer image pyramid model and improved camera pose estimation process
Fig. 3. Unary edge and binary edge models in g2o library. (a) Unary edge model; (b) binary edge model
Fig. 4. Comparison between ground truths and real trajectories of cameras generated by improved ORB-SLAM2 system on four frame sequences. (a) fr1/desk frame sequence; (b) fr1/desk2 frame sequence; (c) fr1/plant frame sequence; (d) fr1/xyz frame sequence
Fig. 5. Reconstruction results and camera trajectories on four frame sequences by improved ORB-SLAM2 system on ICL-NUIM benchmark datasets. (a) lr_kt0; (b) lr_kt1; (c) lr_kt12; (d) lr_kt3
Fig. 6. Comparison of reconstruction results of Kintinuous, ElasticFusion, and improved ORB-SLAM2 systems on closed-loop region of lr_kt3 frame sequence. (a) Kintinuous system; (b) ElasticFusion system; (c) improved ORB-SLAM2 system
Sequence | Relative path /m |
---|
DVO SLAM | Kintinuous | ElasticFusion | ORB-SLAM2 | Improved system |
---|
fr1/desk | 0.022 | 0.142 | 0.022 | 0.016 | 0.016 | fr1/plant | 0.027 | 0.059 | 0.043 | 0.016 | 0.014 | fr1/teddy | 0.049 | 0.237 | 0.091 | 0.035 | 0.056 | fr1/room | 0.064 | 0.182 | 0.198 | 0.059 | 0.056 | fr1/360 | 0.074 | 0.202 | 0.270 | failed | 0.129 | fr1/desk2 | 0.035 | 0.140 | 0.058 | 0.031 | 0.025 | fr1/floor | 0.035 | 0.140 | 0.058 | 0.031 | 0.035 | fr1/rpy | 0.022 | 0.041 | 0.037 | 0.064 | 0.023 | fr1/xyz | 0.013 | 0.021 | 0.014 | 0.009 | 0.009 |
|
Table 1. Comparison of camera trajectories on TUM RGB-D benchmark datasets
Sequence | Frame | Length /m | Withloop |
---|
lr_kt0 | 1510 | 6.54 | No | lr_kt1 | 967 | 2.05 | No | lr_kt2 | 882 | 8.43 | No | lr_kt3 | 1242 | 11.32 | Yes |
|
Table 2. Details of four frame sequences on ICL-NUIM benchmark datasets
System | 3D model result /m |
---|
lr_kt0 | lr_kt1 | lr_kt2 | lr_kt3 |
---|
Kintinuous | 0.087 | 0.470 | 0.162 | 0.205 | ElasticFusion | 0.058 | 0.505 | 0.185 | 0.219 | ImprovedORB-SLAM2 | 0.007 | 0.467 | 0.155 | 0.190 |
|
Table 3. Comparison of 3D models on ICL-NUIM benchmark datasets