Aiming at the difficulty of establishing accurate mathematical model of camera in explicit non-linear calibration, a new implicit non-linear camera calibration method based on Particle Swarm Optimization (PSO) and Least Square Support Vector Machine (LSSVM) regression was proposed. A least square support vector regression machine was built to exactly approximate to the non-linear imaging relationship between image points and corresponding 3D world coordinates. And PSO algorithm was used to search the optimum parameters of the LSSVM regression model to improve the convergence speed and generalization ability. The calibration results of circular template from standard BP neural network, genetic algorithm, LSSVM and particle swarm optimized LSSVM regression, were compared. The comparison analysis indicates that the proposed LSSVM regression method based on PSO has advantages such as higher accuracy, faster convergence speed and better generalization ability.