Along with the development of deep learning, efforts are being made to grasping with the robot using only the camera. Above all, a lot of research is being done for grasping in an environment where various objects are mixed. To perform grasping in complex environments, it is necessary to train the grasping algorithm with vast amounts of data to ensure its robustness. However, collecting grasping data takes a lot of time and effort. In this paper, we proposed the depth tile that simply describes a complex situation by processing a depth image. Through this, the grasping algorithm can use a light artificial neural network, and training data can be generated automatically without grasping in real-world or simulation to minimize learning data collection costs. Artificial neural network trained through the depth tile can perform grasping with high success rate by estimating the grasping angle, which is less likely to interfere with obstacles. In this paper, the proposed grasping method, through experiments to empty randomly placed objects, is proved to be robust in complex environments.