Bin-Picking system containing of industrial robot and machine vision is one of the research hotspots of intelligent industry. In this paper, based on the application background of the automatic grinding system of steel plates with industrial robot，aiming at common planar objects in industrial environment (e.g. steel plates), the recognition and localization algorithm of planar object based on boundary feature are studied. According to different steel scenario, this paper studied the two contents, one is the study on recognition and localization algorithm of planar object based on boundary feature of depth image，another is the study on recognition and localization algorithm of planar object based on boundary feature of point cloud. In their respective application scenarios, both algorithms realize the identification and precise positioning of the steel plate. In this paper, a method for identifying and locating plate objects based on the edge feature of depth image is proposed. The algorithm is divided into two stages: offline modeling stage and online detection stage. In the off-line modeling stage, the cosine value lookup table and the template to be detected are established. Online detection involves the computing gradient images and the searching based on image pyramids. This paper improves the efficiency of online detection by reducing the number of edge points of the base template, optimizing the rotation Angle and searching strategy based on the pyramid. In addition, the algorithm has been successfully applied to the industrial robot automatic plate grinding system. The statistical results of the system at the scene concluded that not only the positioning accuracy of the method is within the scope of project tolerance, but also the method meets the requirement of the project in real time. For the messy steel plate, on the base of the voting-based pose estimation algorithm, this paper proposed an efficient planar object recognition and localization algorithm based on boundary point pair feature. This method can be divided into two stages, the offline modeling stage and the on-line identification stage. In the offline modeling phase, the edge points are extracted for all steel plates in the first place, and then point pair feature is established for every edge point that satisfies the constraint conditions, and the point pair feature with the point pair is stored in a hash table. In the on-line identification stage, plane segmentation is carried out on the scene in the first place, and then edge points are extracted for every plane point cloud. The point pair features are established for every reference point with each other edge point using the same feature description in the offline modeling phase. And then the feature alignment and pose estimation are performed. In addition, the accuracy and precision of the matching are further improved by using the pose clustering and ICP pose optimization in the matching process. In this paper, the experimental platform was set up in the laboratory, and a large number of test experiments were carried out to verify the effectiveness of the algorithm on the identification and positioning of the disorganized object with certain inclination Angle.