Abstract: Visual servoing referred to the technique which uses the computer vision data to control the motion of a robot. Among its various schemes, the image-based method was popular because of many advantages. But there were still some problems in its practical applications. Recently, a general control law of the image-based method using a behavior parameter had been proposed to improve the performance of the method. This study deals with the camera retreat problem and the image-based visual servoing scheme. The behavior-based visual servoing scheme was utilized. An analytical study of this scheme was presented when camera moves along and around the optical axis. The range of behavior parameter was obtained through the analytical study and rbf network. The camera retreat would not occur if the behavior parameter was picked in the range. When considering a difficult configuration, the study proposed an approach to adaptively select behavior parameter according to the range. Simulation results showed that this method compared with the classical image-base visual servoing schemes had better performance in avoiding the local minima in some configurations.