Aiming at improving the low efficiency in CNC machine tool coaxial testing, a new type of CNC machine tool coaxial testing technology is developed through visual imaging system to build a mechanical control system that can meet the automatic positioning of feed system of CNC machine tool. Research results show that when the camera pixel is 5 million, the average error of the arc piecewise circular algorithm is 0.046 mm, and the total time required is 242 ms to achieve the real-time display effect. Moreover, fast arc detection algorithm can achieve high accuracy in a short time under the condition of low camera resolution, so as to improve the final image processing accuracy. Furtermore, the application test shows that the response time required by this algorithm is 244 ms, and it can control the chuck offset in real time. The average error is 0.038 mm, the maximum is 0.075 mm, and the alignment time of CNC machine tool is greatly shortened without decreasing the accuracy. This research has a practical significance for improving the machining accuracy of CNC machine tools, which is easy to be applied widely.