Visual tracking method of metal mold combining local directional feature and association matrix

To solve the problem of positioning the metal mold which contains metals in different states during the casting process, a tracking method combining local directional feature and data association matrix is proposed. Firstly, the local visual directional characteristics of the mold are extracted from multiple mold images. Secondly, the mold template is established from the local directional features, and molds can be positioned through matching. To achieve dynamic tracking of multiple molds, the idea of deepsort is used to build a multi-target detection framework for molds. The association matrix of the mold is obtained by the frames correlation, and the mold ID is marked and tracked according to the association matrix. Finally, the trajectory of the mold is mapped to a three-dimensional space. Experimental results show that the proposed method can be used to detect multiple molds and extract the running trajectory of molds at the same time. The method has advantages in accuracy, reliability and real-time performance, and can provide technical information for the industrial robots.