Abstract: Visual measurement system based on industrial robot is applied to the online and real-time monitoring and inspection of key dimensions in the process of automobile body assembly and subassembly.Temperature variations of the manipulator and environment lead to expansion or deformation of links and joints,therefore cause the temperature drift error. An online calibration approach is proposed.Firstly,positioning error model of the end-effector is established by adopting D-H forward kinematics and differential kinematics.Secondly,combining with thermal experiments on axis rotation,link parameters that change significantly because of robot's temperature variation are determined by using multivariate Unear regression method.At last,thermal error is compensated by using a reference sphere.Practical measurement results show that the measurement errors caused by temperature variation can be reduced to about 0.05 mm and the approach can meet the requirement of the production tact.