A Robust to Distractor Detection Method for Multi-layer Industrial Parts Based on Stereo Vision

In robot grasping tasks, it is a challenge to achieve rapid and accurate object detection. Particularly, the layout of industrial parts is multi-layer, multi-type and disordered (see Fig. 1), so the object detection only relying on RGB image method does not perform well. In this paper, we propose to apply stereo vision combined with convolutional neural network to achieve precise identification and positioning of the multi-layer industrial part. In order to improve the detection accuracy, we use point cloud transformation in stereo vision to preprocess the obtained image information, which takes full advantage of the spatial structure relationship between each object. Based on the historical data of the research objects, the results show that our method can improve the recognition rate of multi-layer industrial parts than the traditional RGB object detection algorithm under the same conditions.