In recent years, researches on autonomous mobile robots are being advanced, and among them, practical use of automatic driving cars is expected particularly. In this realization, there is a parking problem as one of the problems to be solved. Currently, although various sensors such as infrared lasers and millimeter wave radars are used for parking problems, it is desirable to use the less number of sensors and select cheaper ones when considering the cost. This research aims to recognize the surrounding environment for parking control by only images obtained from using a monocular camera. For safe parking, it needs to detect the parking position and check the space to turn the wheel. A technique called Convolutional Neural Network (CNN), which is now a mainstream in image recognition fields, is used to archive them. This paper constructs a system for judging a parking possibility using object detection and a system for classifying several patterns of the space to turn the wheel using depth images. Some simulations are conducted to verify whether the surrounding environment in parking can be integratedly judged by the proposed system.