Research on Target Detection and Grabbing Positioning Method Based on Deep Learning

For livestock and poultry carcass segmentation, the labor intensity is high, the skill requirement is high, and the segmentation efficiency is low; The segmentation process of livestock and poultry meat takes a long time, which is easy to cause secondary pollution of meat and other problems. It is necessary to upgrade the existing livestock meat processing industry and improve the intelligence degree of the current livestock carcass processing technology. In this paper, a fast adaptive histogram equalization method is proposed to enhance the contrast of the pig leg image captured by 3D camera, and then the trained YOLOV4 model is used to detect and calibrate the specified bone position of the pig leg image. This method first needs to establish local coordinate systems for the manipulator and the grasping target respectively, then uses Eye to Hand hand calibration system to obtain the corresponding conversion matrix, and transmits the position information of the grasping target to the manipulator, so that the manipulator can move to the designated position correctly and complete the grasping operation successfully. Field experiments verify the feasibility of this method.