Research On Object Detection And Tracking Method For Family Service Robot

At present,family service robots are a hot research topic in the field of robotics all over the world,which showing a wide range of application prospects.Object detection and tracking is an important part of family service robots.It covers the theoretical knowledge of image processing,artificial intelligence,sensors and other disciplines.It is widely used in intelligent monitoring,visual navigation,human-computer interaction,home service and other fields.In the indoor environment of family,the target object becomes diversified,complexity and unknown.There are many problems in detecting and tracking the target object steadily in the complex environment.Based on family service robots,the paper studies object detection and tracking.The convolution neural network method in deep learning is used in object detection and tracking.The method inputs images or videos into convolution neural network,and extract high-level abstract features from layer by layer.The method realizes multi-target objects detection of indoor objects and tracking target object.The paper focuses on the object detection and tracking of family service robots.Main contents of the paper include:(1)Method and model selection:As for object detection and tracking for family service robots,research studies existing methods.The convolution neural network method in deep learning is applied to object detection and tracking experiments.In order to carry out the next experiment smoothly,the convolution network model,activation function and deep learning framework are selected appropriately according to the hardware equipment and experimental results.(2)Database creation:In order to make the database serve the family service robot better,the existing popular detection and tracking data sets are analyzed.Due to the lack of database related to family indoor environment,we are ready to create Robot-database initially.When collecting database,we need to consider light,occlusion,morphological changes and other factors in the indoor environment in order to improve the training effect of the network.(3)Object detection experiment:In order to achieve multi-target objects tracking in indoor environment,the convolution neural network method is used for object tracking experiment.The convolution neural network is trained by using the Robot-database.The convolution network can learn high-dimensional abstract features independently,which paves the way for object recognition and detection in indoor environment.Because there are certain requirements for the accuracy and real-time performance of object detection,the object detection network will adopt multi-scale feature fusion,full convolution network to extract candidate frames,sharing convolution layer and other methods to optimize and improve the experiment.(4)Intelligent target object tracking:In order to achieve a stable tracking of the target object,the convolution neural network method is used in tracking experiment.Training the tracking network can extract the relationship between the target object motion and appearance.Based on the project background of NEU's first generation family service robot,the robot needs to realize the detection of indoor objects.According to the actual task requirements,the robot also needs to find the target objects and complete the follow-up tracking of the target objects.Lastly,Intelligent target object tracking is realized by combining object detection and tracking.