| Service robots are kind of robot system which can be semi-autonomous or fully autonomous to help humans solve their daily lives.With the development of technology,as nev problems like rising labor costs and aging of the population appear,service robots have gradually become a popular direction and been received a high attention in academia and engineering structure."National Medium and Long-Term Science and Technology Development Plan(2006-2020)"points out that intelligent service robots are the cutting edge technology and focus areas in the next 15 years,which has the characteristic of intellectualization?standardization and systematism.In the service robot system,the visual sensor is a widely used passive sensor,it provides powerful information,and it is easy to use.further study the visual images,eliminates distractions and make them fit the needs of typical applications such as understanding and positioning navigation,has become one of the important research topics of service robots.This paper focuses on the study on the mechanism of visual attention and takes the information perception and application of the service robot as the subject content for research.Set out from the mechanism of visual attention,firstly the general visual attention model is researched.Then,with the carrier of indoor scene recognition,robot navigation,localization,the research work emphasizes model construction,testing using picture databases and real robot platform.The paper specifically includes the following:(1)Based on the analysis of the mechanism of visual attention,existing "bottom-up"models are studied deeply,and a visual attention model based on global statistics and surrounding contrast information is proposed.Among them,the global statistical method adopts auto-threshold value technology,which based on information entropy,to smooth the images.Therefore,while preserving the necessary details of the image,a similar background area is effectively merged.Based on the color feature and the statistical method,the global significant value is calculated on the opposing color space,and the full resolution scene is obtained.For the surrounding contrast model,a widely-used superpixel method has been used to over-segment the input image into small regions,and then the saliency values have been calculated taking the color,spatial and textural distinctness factors into account.The final saliency map has been obtained by integrating the two saliency maps,and the experimental results of fifteen state-of-the-art methods have been compared for three datasets,and have shown that our method can achieve significantly better saliency results in quantitative and qualitative analysis.(2)Conventional scene recognition methods have poor performance in indoor situations.For this reason,a novel indoor scene recognition method,combining global and saliency region f'eatures,is presented.In addition to the use of an improved BoW(Bag-of-Words)model,an improved BDBN(bilinear deep belief network)is implemented,using information from a salient region detection technique.The salient regions with the visual attention approach are sent into the improved BDBN(Bilinear deep belief network)to automatically construct models for image classes.The final result of the indoor scene recognition can be obtained by combining the above-mentioned two models through strategies for a Piecewise discriminant.The evaluation was performed on the real mobile robot platform and the standard MIT 67-category indoor scene dataset.The experiments show that the proposed method is highly effective.and can improve the accuracy of common BoW-based methods by up to 10%.In addition,the accuracy rate of our method represents a competitive advantage over other methods such as SP-Bow,R-BoW,and so on.(3)To resolve the problems of natural landmark-based navigation such as computational complexity and bad robustness,a robot localization and navigation architecture using stable natural landmarks are mentioned.According to the essence of the architecture requirements,a visual attention model in the frequency domain is designed.Then,the saliency map generated by the model is used to guide the feature operator to focus on the fixed area,and this practice can drastically reduce the computational effort of feature extraction and matching.After that,the local descriptors are extracted and are applied to create the topical map.The experimental results indicate that the proposed method does not need camera calibration,and can locate and navigate for service robot,and has the robustness of dynamic information of environment.(4)Sliding-window obj ect detection is a popular technique for identifying and localizing objects in an image.Considering the shortage of huge calculation and low efficiency,this paper creates a objectness model based on the color and super-pixel features.Firstly,the saliency map is computing by the saliency computation method.Meanwhile,the color image is coarsely segmented into several binary images.which is accomplished based on a single refinement skeleton extraction algorithm.Then,using the FCM(fuzzy C-mean)clustering method to improve the image segmentation result.Finally,taking advantage of data fusion technique,the segmented images,and the saliency map can be combined to get the final salient region segmentation results.Combined with the super-pixel features of the image,the maximum density value clustering and hierarchical grouping strategy are designed to further extract the possible obj ect regions in the image.Experiments show that the method can reflect the semantic features of the image,and can get a better segmentation effect.(5)Based on the friendly human-computer interaction perspective,a robot navigation architecture based on inaccurate map is proposed.Firstly,the user should provide object words and their relative positions,and give a hand-drawn path to guide the robot move from the starting point to the destination.In the target identification stage,the visual system uses the objectness measure of image windows based on the color and super pixel features to filter out the possible object regions and identify the target objects in these areas.Comparing with other object recognition algorithms,this method can reduce the number of windows and improve the efficiency.In the navigation stage,the GMS(Grid-based Motion Statistics)algorithm is also introduced to accelerate and optimize the feature matching,which is used to guide the steering of the robot.On the other hand,the ranging sensors are used to assist the robot in obstacle avoidance operation.Compared with the traditional robot navigation methods,the proposed navigation framework has some advantages such as simple and easy to use,low-cost,and can effectively guide the robot to run smoothly in the indoor environment. |
PDF Full Text Request