The absolute positioning precision of industrial robots currently falls significantly low than its repeatability precision. This paper investigates the theory of kinematic parameter errors of robots, explores pose measurement techniques based on binocular vision, and proposes a three-step method for identifying kinematic parameters, with aim at enhancing precision of robot’s TCP (Tool Center Point). The three-step comprises: Firstly, fit the 1st, 2nd, and 3rd joint axes via single-axis rotation to calculate the first set of kinematic parameters; Secondly, based on binocular vision measurement of the target ball's pose, construct error model of the target ball's pose to identify the angular components of the remaining kinematic parameters; Thirdly, construct error model of the target ball's centre position, and identify the length components of the remaining kinematic parameters. This paper constructs an experimental system, and conducts visual measurement research and algorithmic implementation regarding the variation pattern of the ball pose during single-axis rotational motion of robot. The proposed three-step method circumvents the issue of low accuracy in the rear three axes. It separates the angular and length parameters of the kinematics. Experimental validation demonstrates that this proposed compensation algorithm reduces positioning error down to 66.00% of the original and uncertainty down to 85.62% of the original. This approach effectively minimises positioning errors in industrial robots. Comparative analysis of compensation outcomes across one-step, two-step axis, two-step, and three-step methods confirms that the proposed three-step method achieves more efficient identifying of kinematic parameters.