Research On The Method Of Object Detection Of Personal Care Robot Based On Convolution Neural Network

Because the old people are becoming more and more and the living standard of people has been greatly developed,the personal care robots are being got much more attention.In this thesis,we research on the vision module of the personal care robot.To make the detection of robot be real-time,we adopt the object detection algorithm based on convolutional neural network.With the development of artificial intelligence,this method is the main research contents of object detection in robot vision.In this thesis,because of the shortcomings of traditional machine vision object detection algorithm and considering the characteristics of the region convolutional neural network object detection algorithm,we propose an improved faster region convolutional neural network(Faster-RCNN)object detection algorithm,which reduces the time of training and detection.And the improved algorithm is used in personal care robot.Because most of the time of object detection is wasted on feature extraction,this paper firstly needs to find an feature extraction algorithm with simple structure and less detection time,whose detection accuracy could meet the requirement at the same time.Firstly,the MobileNet is compared with other feature extraction algorithms and the experiment result shows that the MobileNet has a better performance,which could reduce the detection time to 119 ms when the accuracy is 76.8%.Then the feature extraction algorithms are applied to Faster-RCNN to carry out the experiments.What's more,during the training,data set labeled by hand is added to make the service robot could have a better performance in more conditions and make the algorithm be more robust.The experimental results show that the combination of MobileNet and Faster-RCNN achieves the best effect,whose training time and detecting time are less than two other algorithms.In the allowable range of detection accuracy,in this thesis,the detection time of Faster-RCNN is reduced to 134 ms when the detection accuracy is 85.2%.Finally,the algorithm is applied to personal care robot for real-world object detection.Under different conditions,the experiment of detecting different objects is carried out.After the improved algorithm is used in the module of robot vision,the robot can perform the detection task accurately with a detection percentage of 100% without interference when the distance varies.Under the condition of weak interference,such as partial occlusion and the reduction of illumination,the personal care robot can still have a good performance on the object detection with a detection percentage of above 90%.To the mission of detecting several objects at the same time,the personal care robot could detect the objects when the objects stay in a proper distance with a detection percentage of 100%.The detection algorithm in vision module of personal care robot is improved whose detection time is reduced.What's more,the robot could detect the objects immediately under complicated conditions.