Research On Garbage And Driving Area Detection Of Sweeping Robot

With the development of science and technology,home service robots,which are closely related to human daily life and work,have attracted more and more attention.In recent years,intelligent sweeping robots have been industrialized on a large scale in the domestic and foreign consumer markets,bringing great convenience to people's lives.At present,the sweeping robot on the market has simple functions such as path planning,automatic charging,operation planning and obstacle avoidance.But the intelligent degree of sweeping robot is not high,and it generally lacks the ability to understand the environment,The ability of complex perception,cognition and decision-making is weak,and the ability to classify and identify garbage is not enough,Cleaning efficiency is low and path planning method with strong adaptability is lacking.The environmental perception ability of the sweeping robot based on machine vision is studied,and the garbage is accurately detected and classified,It is beneficial to enhance the initiative and efficiency of the sweeping robot in garbage cleaning,At the same time,it can greatly enhance the ability of the sweeping robot to classify and dispose of different garbage.In order to improve the intelligence of the sweeping robot,this paper focuses on garbage detection and classification,and the detection of driving area.The paper mainly completes the following three parts:(1)Build a test platform of sweeping robot,The platform includes embedded vision processing system,vehicle camera,omnidirectional motion Anycbot four-wheel drive robot chassis.Equipped with visual sensors to support garbage detection and classification and driving area detection,It meets the requirements of principle verification and algorithm verification.(2)In view of the problem of fast and accurate detection and classification of garbage in home environment,YOLOv2 network with fast detection speed is chosen as the main network.At the same time,in order to make full use of the high resolution features of the image,Combining YOLOv2 network with densely connected convolutional network,The shallow and deep features of the image are multiplexed and fused by embedding deep dense modules,It can reduce the loss of feature information.Finally,the training of garbage detection and classification model is completed by using training data enhancement processing and multi-scale training strategy.Experiments show that the accuracy of garbage detection and recognitionreaches 84.98%,and the real-time detection speed reaches 26 frames per second,which basically meets the requirements of real-time and accuracy.(3)In order to detect the moving area of the sweeping robot in indoor scene,a deepabV3+ network model for semantic segmentation is selected to discriminate the moving area of the sweeping robot.deepabV3+ network model using encoder-decoder architecture.Encoder uses Xception model and empty space pyramid pooling to capture higher semantic information and increase the ability of feature extraction,Decoder uses a simple and effective module to gradually restore the target details and corresponding spatial dimensions.This method has been tested on the open ADE20 K data set and the built data set,and achieved good results.Although the segmentation effect on multi-target segmentation task needs to be strengthened,it But it solves the problem of indoor driving area segmentation well.The attempt made in this paper on the detection and classification of garbage and the detection of driving areas in indoor scenes will help to enhance the intelligent level of the sweeping robot.The research methods have important reference value for the research and development of intelligent sweeping robot technology,and the related methods are worthy of further research and analysis.