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, Shenyang, Liaoning 110016, China5 Key Laboratory of Image Understanding and Computer Vision, Shenyang, Liaoning 110016, Chinashow less
Fig. 1. Segmentation results obtained with two algorithms. (a) 3D point cloud (top view); (b) segmentation result obtained with original RANSAC algorithm; (c) segmentation result obtained with modified algorithm
Fig. 2. Results obtained before and after region merging. (a) Connected planar regions; (b) merged result obtained with glue algorithm
Fig. 3. Intermediate results obtained with proposed algorithm. (a) Connected planar regions; (b) merged result obtained with glue algorithm; (c) binary image corresponding to the red connected region; (d) binary image after erosion; (e) final segmentation result
Fig. 4. Plate-shaped objects and data acquisition platform. (a) Six types of plate-shaped objects; (b) simple scene; (c) complex scene
Fig. 5. An illustration of depth image and its corresponding three-dimensional point cloud. (a) Depth image; (b) three-dimensional point cloud (top view); (c) amplified three-dimensional point cloud
Fig. 6. Segmentation results obtained with four algorithms. (a) Depth images; (b) Canny edge algorithm; (c) K-means algorithm; (d) RANSAC algorithm; (e) region growing algorithm
Fig. 7. Comparison between original depth image and labeled images. (a) Original depth image; (b) ground truth (gray image); (c) ground truth (color image)
Fig. 8. Segmentation results obtained with proposed algorithm on twenty simple scenes
Fig. 9. Segmentation results obtained with two algorithms on S6, S8 and S15 scenes.(a) Proposed method; (b) region growing algorithm
Fig. 10. Segmentation results obtained with two algorithms on ten complex scenes. (a) Proposed method; (b) region growing algorithm
Nr | τransac /mm | Np | Ne | Ns | α /(°) | Nm | Nc |
---|
1000 | 3.5 | 300 | 50 | 500 | 80 | 1.5Mmin | 100 |
|
Table 1. Parameter setting of proposed algorithm
Scene No. | Segmentation accuracy on simple scenes | Segmentation accuracy on complex scenes |
---|
Region growing algorithm | Proposed algorithm | Region growing algorithm | Proposed algorithm |
---|
S1 | 1 | 0.99 | 0.95 | 0.96 | S2 | 0.94 | 0.94 | 0.85 | 0.93 | S3 | 0.97 | 0.96 | 0.97 | 0.97 | S4 | 0.91 | 0.94 | 0.92 | 0.89 | S5 | 0.96 | 0.94 | 0.96 | 0.92 | S6 | 0.82 | 0.89 | 0.96 | 0.96 | S7 | 0.90 | 0.91 | 0.93 | 0.96 | S8 | 0.84 | 0.99 | 0.84 | 0.95 | S9 | 0.99 | 0.99 | 0.93 | 0.94 | S10 | 1 | 0.99 | 0.92 | 0.95 | S11 | 0.97 | 0.97 | | | S12 | 1 | 0.99 | | | S13 | 1 | 0.99 | | | S14 | 1 | 1 | | | S15 | 0.94 | 0.95 | | | S16 | 1 | 0.95 | | | S17 | 1 | 1 | | | S18 | 1 | 1 | | | S19 | 1 | 1 | | | S20 | 0.98 | 0.98 | | | Average | 0.961 | 0.968 | 0.923 | 0.943 |
|
Table 2. Segmentation accuracy obtained with proposed algorithm and region growing algorithm on twenty simple scenes and ten complex scenes
Image size /(pixel×pixel) | Average time cost on simple scenes /s | Average time cost on complex scenes /s |
---|
Region growing algorithm | Proposed algorithm | Region growing algorithm | Proposed algorithm |
---|
640×480 | 11.02 | 2.14 | 11.89 | 3.14 | 320×240 | 2.65 | 0.57 | 2.59 | 1.02 |
|
Table 3. Average time cost obtained with proposed algorithm and region growing algorithm on twenty simple scenes and ten complex scenes