Research Of Monocular Camera Calibration Algorithm Based On Particle Swarm Optimization

Computer vision has the advantages of low-cost, non-contact, as well ashigh-precision. So it is widely used in industrial measurement, video monitoring,medical imaging, and other areas. However, the accuracy of computer visionmeasurement system largely depends on the camera calibration accuracy. Therefore,the research on the methods of camera calibration has become one of the hot fields in theworld in recent years. In order to improve the calibration accuracy, the parameters of thetraditional camera calibration methods must be optimized. Traditional optimizationalgorithms, such as LM method and Newton’s method, strongly depend on the initialvalue, have the poor convergence performances, and are hard to get the global optimalresult. Moreover, the traditional methods need to calculate the gradients of themulti-peak equations, thereby being complex and time-consuming.To solve the above mentioned problems, the thesis has investigated theoptimization algorithms for the monocular camera calibration. The thesis presents themonocular camera calibration algorithm based on the Standard Particle SwarmOptimization (PSO) algorithm and Quantum Particle Swarm Optimization (QPSO)algorithm. In short, the following research work has been accomplished.(1) Extraction algorithm for the target feature points. According to imagingcharacteristics of the circular targets, this thesis converts the extraction of circulartargets to the extraction of ellipse targets. An approach for pixel edge points detectionbased on gray difference centroid algorithm is proposed firstly. Then the Zernikemoments method is used to compensate the sub-pixel edge points. At last, the accurateellipse’ center is acquired by the algorithm of ellipse fitting to the edge points.Experimental results indicate that with the proposed algorithm, the maximum error ofellipse fitting is less than0.08pixels, and the average error is about0.02pixels. Theproposed algorithm has a high localization precision.(2) Camera calibration algorithm based on PSO. Traditional camera calibrationmethods usually employ nonlinear optimization algorithms, such as LM method,Newton’s method. The main drawback of them is easy to converge to a local optimalpoint. Therefore, this thesis proposes an approach to get the accurate parameters forcamera calibration based on PSO algorithm. The thesis first uses the sum of areasquare as a new distortion measurement index, thus obtaining more accurate values for distortion parameters. Using the back-projection error as the fitness function, theproposed PSO based calibration algorithm updates the velocity vector and positionvector of each particle iteratively until the global optimal position is obtained.Experimental results demonstrate that the proposed calibration algorithm canaccomplish the calibration process reliably, and the accuracy of the optimized camerainner parameters can satisfy the practical calibration requirement.(3) Camera calibration algorithm based on QPSO algorithm. In order toovercome the local optimization problem of the standard PSO algorithm, this thesisalso investigated the camera calibration algorithm based on QPSO algorithm.Different from PSO, the evolutionary optimizing process is implemented by updatingthe position vector from the quantum behavior of particle. Hence, the optimal processcan has higher probability to converge to the global optimal position. Experimentalresults show that the QPSO algorithm has faster convergence speed, biggerprobability to get the global optimal position vector than the standard PSO algorithm,and can improve the accuracy of optimal result effectively.