Stator wedge tightness inspection is one of the most important periodic inspections carried out on required for large-capacity generators. The traditional wedge tightness inspection method enforces inspectors to enter the narrow and uncomfortable inner space of a stator. As an alternative, a wedge tightness inspection mobile robot is developed. This robot has a robotic arm with a wedge tightness test module. A visual navigation system using two cameras is proposed. The proposed system controls the driving of the robot in a stator by using line detection of the bottom view camera image. The system also detects wedges and assigns a location number to them by using the inspection view camera image and an object detector based on a convolutional neural network. The detected wedges are tracked using robot motion to assign location numbers to the new wedges. The inspection points are then determined using the length and position of the detected wedges. The experimental results conducted on the stator testbed demonstrate the potential applicability of the proposed system.