An autonomous mobile robot has to possess a locomotion way suitable for the mobile environment and to autonomously move. This research aims to operate an autonomous mobile robot by determining the control input from the features of acquired camera images. Some advantages of this method are the following points: i.e., there is no need for self-localization, so that the amount of computations is small, and it can be controlled by a human-like method. In conventional studies, avoiding of static obstacles was realized by using edge detection and fuzzy control. However, there remain several problems, i.e., it is difficult to deal with moving obstacle, there is no system for approaching a target point while avoiding obstacles, etc. Therefore, this paper proposes a method for approaching a target point while avoiding obstacles, by adding potentials around the target when constructing the potential field. In addition, a collision prediction system is constructed using a convolutional neural network (CNN) and a method of stopping a robot in the vicinity of the target point while avoiding a moving obstacle is proposed by switching control based on such a system. The effectiveness of the proposed method is verified by simulation experiments. The experimental results show that the robot avoids a moving obstacle and stops at the target by using the proposed method.