Binocular measurement model of locating fruit based on neural network

Aiming at the problem of locating fruit by using binocular system under natural condition, an implicit 3D measurement model of binocular system based on neural network was proposed to locate fruit. The model used two Ward network to associated work and simulate geometric localization model of binocular system based on parallax computation. The method needn´t take account into some localization procedures including image correction and error modification. The results showed that maximum absolute error in horizontal, vertical and depth direction of trained model is less than 10mm in effect viewing field of 223 mm to 1183 mm. Positioning reliability can reach 88.584 percent when resultant error in horizontal, vertical and depth direction is limited within α5mm. Real tests of locating tomato showed that the precision of localization is acceptable.