Research And Implementation Of Complex Object Detection Based On Cloud Robotic Architecture

Robot object detection refers to the process of locating the target object in the range of the sensor based on the given name or characteristic of the target object by means of machine vision or other methods.It is a basic enabling technology for robots to do handling,searching,alerting and other tasks,which are of crucial importance in the field of robots.At present,the detection of robotic objects mainly aims at detecting objects described by simple labels(such as "coffee cup"),through the means of image recognition and matching.In many real-life scenes,object detection instructions are often described by complex objects that consists of a set of tags and modifiers,such as "a gray coffee cup next to a television cabinet with a spoon on it".The existing methods based on image recognition face challenges and are difficult to meet the needs of real applications because of the dynamic,variability and uncertainty of this complicated semantic description.In this paper,the task of detecting complex semantic descriptions would be called complex object detection.Complex object detection in the real world is a typical data-intensive task that requires a great deal of knowledge to support.At the same time,the object detection algorithm based on deep learning also needs to consume a large amount of computing resources.Cloud robots are expected to provide effective support for such data and intensive tasks.In this paper,the detection of complex objects based on cloud robots is taken as the research object.Two types of complex objects,which are single objects and multi-modified objects with multiple modification attributes,are explored as follows:(1)A complex object detection method based on active image synthesis is proposed.This method aims at single object detection tasks with multiple modification properties.The possible target images would be synthesized through the generation of confrontation network,and then matches and detects the synthesized images with the images collected by robots in the real scene.Experimental results show that the mAP measured by this method on some CUB datasets is about 60.83%,which is significantly improved compared with the traditional object detection algorithm.(2)A method of complex object detection based on image semantic analysis is proposed.The method takes the object detection with multimodality as the object,semantically decomposes the objects and their combined relations in the current machine vision images,and then matches the semantic information with the complex semantic description.Experiments show that this method can better deal with the detection problem of combinatorial objects with multiple modification attributes,and the mAP reached 68.20%on the test of VOC dataset(3)Proposed the capability expansion mechanism of the above two methods in cloud robot architecture.Based on the architecture of cloud robot,this paper proposes a computational unloading mechanism of the above method and an online method of online detection based on internet data in the detection process.On the basis of the above methods,this paper realizes a prototype detection system for complex objects of cloud robots.Experiments on open datasets and data collected from real scenes are carried out,and the work is fully verified.