<abstract xmlns="http://www.w3.org/1999/xhtml"><p>A new approach based on hybrid Hopfield neural network and self-adaptive genetic algorithm for camera calibration is proposed. First, a Hopfield network based on dynamics is structured according to the normal equation obtained from experiment data. The network has 11 neurons, its weights are elements of the symmetrical matrix of the normal equation and keep invariable, whose input vector is corresponding to the right term of normal equation, and its output signals are corresponding to the fitting coefficients of the camera’s projection matrix. At the same time an innovative genetic algorithm is presented to get the global optimization solution, where the cross-over probability and mutation probability are tuned self-adaptively according to the evolution speed factor in longitudinal direction and the aggregation degree factor in lateral direction, respectively. When the system comes to global equilibrium state, the camera’s projection matrix is estimated from the output vector of the Hopfield network, so the camera calibration is completed. Finally, the precision analysis is carried out, which demonstrates that, as opposed to the existing methods, such as Faugeras’s, the proposed approach has high precision, and provides a new scheme for machine vision system and precision manufacture.</p></abstract>

A new approach based on hybrid Hopfield neural network and self-adaptive genetic algorithm for camera calibration is proposed. First, a Hopfield network based on dynamics is structured according to the normal equation obtained from experiment data. The network has 11 neurons, its weights are elements of the symmetrical matrix of the normal equation and keep invariable, whose input vector is corresponding to the right term of normal equation, and its output signals are corresponding to the fitting coefficients of the camera’s projection matrix. At the same time an innovative genetic algorithm is presented to get the global optimization solution, where the cross-over probability and mutation probability are tuned self-adaptively according to the evolution speed factor in longitudinal direction and the aggregation degree factor in lateral direction, respectively. When the system comes to global equilibrium state, the camera’s projection matrix is estimated from the output vector of the Hopfield network, so the camera calibration is completed. Finally, the precision analysis is carried out, which demonstrates that, as opposed to the existing methods, such as Faugeras’s, the proposed approach has high precision, and provides a new scheme for machine vision system and precision manufacture.

Camera Calibration by Hybrid Hopfield Network and Self- Adaptive Genetic Algorithm

Department of Mechanical and Energy Engineering, Shaoyang University, Daxiang District, Shaoyang, 422004, P. R. China

College of Mechanical and Vehicle Engineering, Hunan University, Yuelu District, Changsha, 410082, P. R. China

School of Mechanical and Automotive Engineering, South China University of Technology, Tianhe District, Guangzhou, 510640, P. R. China

Department of Production Engineering, School of Industrial Engineering and Management, Royal Institute of Technology (KTH), S-100 44, Stockholm, Sweden

Department of Mechanical Engineering, Faculty of Engineering, Kyushu University, 744, Motooka, Nishi-ku, Fukuoka 819-0395, Japan

Measurement Science Review

A new approach based on hybrid Hopfield neural network and self-adaptive genetic algorithm for camera calibration is proposed. First, a Hopfield network based...

Camera Calibration by Hybrid Hopfield Network and Self- Adaptive Genetic Algorithm

Online veröffentlicht: 15. Dez. 2012

Seitenbereich: 302 - 308

DOI: https://doi.org/10.2478/v10048-012-0042-5

Schlüsselwörter
Camera calibration, projective matrix, Hopfield neural network, self-adaptive genetic algorithm, longitudinal direction and lateral direction

This content is open access.

Camera Calibration by Hybrid Hopfield Network and Self- Adaptive Genetic Algorithm

Online veröffentlicht: 15. Dez. 2012

Seitenbereich: 302 - 308

DOI: https://doi.org/10.2478/v10048-012-0042-5

SchlüsselwörterCamera calibration, projective matrix, Hopfield neural network, self-adaptive genetic algorithm, longitudinal direction and lateral direction

This content is open access.

Schlüsselwörter
Camera calibration, projective matrix, Hopfield neural network, self-adaptive genetic algorithm, longitudinal direction and lateral direction