| 摘 要: 在机器人识别并抓取物体中,视觉识别虽取得显著进展,但在视觉图像不足以区分物体时,识别准确率会受到影响。触觉数据能提供重要的互补信息,结合视觉与触觉数据能实现更全面的物体识别。为了有效融合视、触觉信息,提出了一种创新且经济的方法获取触觉数据,设计了多尺度融合机制以应对两种模态的差异。实验结果表明,该融合机制使物体分类准确率达到98%,较未融合网络提升1.73个百分点。结论表明视触融合模型可显著提升物体识别性能。 |
| 关键词: 多模态 信息融合 多尺度 视觉识别 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 国家自然科学基金资助项目(61663005) |
|
| Researchon Object Classification and Identification Based on Vision-Tactile Fusion |
|
YU Hang, LI Handong
|
(College of Electrical Engineering, Guizhou University, Guiyang 550025, China)
2926479961@qq.com; 470394668@qq.com
|
| Abstract: In robot object recognition and grasping, although visual recognition has made significant progress, the accuracy of identification can be compromised when visual images are insufficient to distinguish objects. Tactile data provides crucial complementary information, and combining visual and tactile data enables more comprehensive object recognition. To effectively fuse visua-l tactile information, this paper proposes an innovative and cos-t effective method for acquiring tactile data and designing a multiscale fusion mechanism to address the differences between the two modalities. Experimental results demonstrate that this fusion mechanism achieves an object classification accuracy of 98% , representing a 1.73 percentage points improvement over networks without mult-i modal fusion. The conclusion indicates that the vision-tactile fusion model significantly enhances object recognition performance. |
| Keywords: multimodal information fusion multiscale visual recognition |
