Mobile Robots Environment Recognition Based On Neural Network

Mobile Robots apperceive the outside world mainly through vision, touch and range sensors, among which the vision sensor can obtain most of the information about the environment. It is full of theoretical and realistic significance to research on how to use Mobile Robots' vision sensor for the environment recognition.It is an effective method to image pattern recognition in recent years. Traditionally, image pattern recognition based on neural network needs feature extraction first, and then the most important features extracted by the researchers are delivered to the neural network classifier for training and recognizing. However, there is no unified method to solve the problems such as which features to choose and how many features to maintain.In this paper, a new type of neural network model--PARCONE (PartiallyConnected Neural Evolutionary) proposed by our laboratory was improved, and was applied to the Mobile Robots environment recognition. The method can overcome the disadvantage that the previous neural networks can not accept more than thousands of inputs. In the new model, firstly, all neurons can be connected with a fixed number of other neurons, which we call partially connected. In the process of training, all pixels of the image what Mobile Robots observes are inputted into the network instead of feature extraction. Evolution occurs by mutating the weight values, and by creating and deleting random connections by genetic algorithm (GA). In this way, our neural network can automatically select the most important features of the environment to avoid information loss caused by human selective features. During the period of recognition, all pixels of the environment image are also set as input neurons. Through the network, the output is the result for environment recognition.Using Visual C++ 6.0 as a design platform, the paper researches on vision-guided environment recognition system of Mobile Robots based on PARCONE algorithm, and realized this method on "NAO" robot. The experimental results show that the new neural model has not only a strong recognition ability in environment recognition, but also has good robustness against image transformation such as illumination, rotation and scale transformation. The Mobile Robots using the PARCONE algorithms can recognize environment well.