A self-calibration technique based on genetic algorithms (GAs) with simulated binary crossover (SBX) and laser line imaging is presented. In this technique, the GA determines the vision parameters based on perspective projection geometry. The GA is constructed by means of an objective function, which is deduced from the equations of the laser line projection. To minimize the objective function, the GA performs a recombination of chromosomes through the SBX. This procedure provides the vision parameters, which are represented as chromosomes. The approach of the proposed GA is to achieve calibration and recalibration without external references and physical measurements. Thus, limitations caused by the missing of references are overcome to make self-calibration and three-dimensional (3-D) vision. Therefore, the proposed technique improves the self-calibration obtained by GAs with references. Additionally, 3-D vision is carried out via laser line position and vision parameters. The contribution of the proposed method is elucidated based on the accuracy of the self-calibration, which is performed with GAs.
An automatic technique to provide interpolation and continuity in the object surface modeling via Bezier networks is presented. This technique generates the surface model via Bezier networks based on surface points, which are retrieved via laser line scanning. In this model, a Bezier function is added in the surface edge to provide continuity G1 and interpolation. Preserving these surface properties via network, the model accuracy and object representation are improved. Also, the surface model reduces operations and memory size to generate the object surface. This is because the model is implemented with fewer mathematical terms than the traditional models. The surface model is defined by means of network weights and the control points. The surface measurement is carried out by a Bezier network via line position to avoid external measurement errors. Thus, the computational model generates the object shape with high accuracy. This is because the network interpolates all points of the physical object. The contribution of the proposed technique is corroborated by an evaluation based on interpolation, continuity, memory size, and speed of the traditional models. The evaluation shows evidences of the viability of the proposed network.