Visual Saliency Detection For Welding Area

In the harsh welding field environment,the traditional machine vision system is difficult to determine the target of interest and obtains clear image information facing the complex visual information.In order to make up for the shortage of traditional welding vision sensor and obtain effective information of welding area,in this thesis,based on the visual saliency detection technology,the following work based on the visual saliency detection technology has been done:(1)In order to obtain the information of laser stripe in the welding area,a structure light information acquisition technology based on the saliency of welding area image is proposed.In the welding field,the laser direction of the laser welding vision sensor is basically fixed.According to this feature,the Gabor filter can be used to extract the direction feature map of the image,and the stripe of the laser stripe is significantly enhanced.After the saliency map is normalized as the weighting map of the corresponding pixel weight,the weighted image is weighted to the original image,and the image with clear laser stripe information is outputted.The non-laser stripe background information is suppressed,so as to prepare for the weld seam feature point extraction work.(2)In order to obtain the information of weld pool and laser stripe at the same time,an image acquisition technology based on the visual features of the welding area is proposed.Based on the luminance information characteristics of the weld pool and the laser stripe are significant for brightness measurement feature map,combined the direction saliency map of the welding area image to get a saliency map as the weight map of the original image,the original image and the weight map are weighted normalized,which obtained a image contains a laser stripe and weld pool information.(3)Aiming at the phenomena of strong interference and uneven light in the welding area,a multi-exposure mechanism based on a single light spectrum and image dynamic homogenization compensation technologies are proposed in the welding field.The response curves of the camera and the response of the pixel value to the value of illumination are obtained by the calibration of the response curve of the camera with different exposures time frame.Then an image can be obtained which contains the welding seam and molten pool and the laser fringe information at the same time.(4)In order to obtain the whole image of the welding area,a multi-scale multi-exposure significant image fusion technique is proposed.The human visual system observation objects in different distance will get the characteristic information of the objects under different scale.Based on the characteristics of the human visual system,the salient maps of the original images in different resolution are obtained by using pyramid multi-scale decomposition method.Combined with the image exposure information,an image which contains rich information of the real scene and is more consistent with human visual characteristics is acquired.At the same time,the saliency map of the image under different scales can be obtained by using the technology of spectral residual,which can improve the image saliency and reduce the information redundancy caused by multi-exposure and multi-scale images.