Study on Adaptive and Fuzzy Weighted Image Fusion Based on Wavelet Transform in Trinocular Vision of Picking Robot

Abstract

In order to improve the adaptive recognition abilities of picking robot in complex environment, a fusion approach of trinocular vision in wavelet domain based on fuzzy reasoning weight was proposed. Firstly, membership functions of fusion rules are determined by fuzzy reasoning of picking environmental features, and membership values of fusion types are calculated according to regional energy and match degree of origin images. Based on the maximum membership degree principle, fuzzy decision is carried on to determine the fusion types and fusion weight. Secondly, the mean weighted method and regional energy feature method are adopted respectively to carry on the low frequency as well as high frequency coefficients fusion among multi-source images by using two-level 2D wavelet, and the final fusion images are attained by inverse HIS transform based on inverse wavelet transform. Four groups of experiment show that in the complex picking environment like weak illumination and strong noise, the information entropy and average gradient of fused image that obtained by using the wavelet fusion method based on fuzzy reasoning weight are higher than that of traditional mean method, pyramid algorithm and wavelet packet method, which means that the fusion effect has been improved greatly.