Currently most of the traffic camera self-calibration algorithms are performed based on vanishing point, however, there would be a tending to infinity ill-condition of vanishing point in certain angle of view. To overcome this problem, firstly, we establish a typical complementary calibration models, then we get vanishing points through the diamond space created by the vehicle trajectories and body edges, meanwhile detect geometrical markings in road active area, finally, in order to avoid the “shock” effect of the vanishing point, adopt the optimization method to optimize the parameter space of the self-calibration model by using redundant information to make the model more accurate. The experimental results on real traffic images demonstrate the effectiveness and practicability of our self-calibration method, and it is especially suitable for PTZ cameras that constantly changing angles of view.