Research Of The Robot Target Recognition System In 360°Environment

Images are applied frequently in the engineering and scientific research, which need to be identified. The traditional image recognition technology as one of the core technology which promote the production and living to be automatic, modern and intelligent, the limitations of its scope of recognition can not meet the recognition problem in many areas. In this paper expensive image acquisition devices with wide-angle do not be used, but the computer algorithm is improved and image stitching technology is selected to achieve the wide-angle, and coupled with BP neural network with highly nonlinear mapping ability we can improve recognition reliability easily. This method can provide a great supporting effect for barrier identification and selection of obstacle navigation method. In this paper we make research based on information of target images which obtained from simulation of some common objects. After a series of reference to domestic and foreign advanced technology, we build an image recognition system combined with the practical needs. The innovations are:Firstly, the image stitching technology was used to expand the visual range of image acquisition; this process not only expand the visual range, but also make the image pre-processing which can improve the clarity of the image, which increase the reliability of the subsequent accurate identification of target image. Secondly, the multi-feature is extracted from the stitched image, and then the super-strong nonlinear mapping ability of BP neural network is used to build the mapping between the multi-feature of the target image and obstacle type, ultimately, the pattern recognition has been made for simulated obstacle. After making full analysis about various parameters of the BP neural network, The LM(Levenberg-Marquardt)optimization algorithm is cited to replace the traditional steepest gradient descent method as the BP neural network training function. It is obviously from the final test process, the improved algorithm has been greatly improved from the speed and accuracy.Finally , MATLAB and LabVIEW are used to make co-simulation, building a complete man-machine system for 360°environment target identification, where play the LabVIEW's advantages which are simplified procedures and short development period and MATLAB's powerful functions are the matrix operations, image processing and numerical analysis.