Resarch On Real-time Target Detection And Attitude Recognition Based On ROS Robot

In recent years,with the development of deep learning technology,using deep learning to solve some problems in the field of computer vision has become the current mainstream research direction.Compared with traditional image processing methods,using deep learning can get higher accuracy and the process is simpler.However,some deep learning models are less accurate or less real-time when ported to mobile devices.Aiming at the practical application scenario of ROS mobile robots,this paper conducts researches on object detection and pose recognition based on deep learning.The main work is as follows:In view of the low accuracy of the original YOLOv5 s object detection model,this paper first improves the backbone network of YOLOv5 s to obtain YOLOv5-light.On this basis,in order to further improve the accuracy of the improved network,the attention mechanism module is added to obtain YOLOv5-light-attention.The performance of the improved model was verified on COCO dataset,and the results show that the detection accuracy of the improved model is 3.2% higher than original YOLOv5 s model.Finally,an object detection model of detection handle was obtained by using the improved model on the self-built data set,and verified on the test set.Experimental results show that the accuracy of the improved model is 90.2%,and the real-time performance is strong,which is suitable for the application of mobile robots.To solve the problem of slow inference speed in the original Open Pose pose estimation model,this paper first makes lightweight improvement on the Open Pose network,and then uses the tag fusion correction method to further improve the accuracy of the model.The performance of the improved model is verified on COCO dataset,and the results show that the accuracy of the improved model is not much different from the original Open Pose model,but the detection speed is increased by 4 times.Finally,a pose recognition model is trained on the self-built data set by using the skeleton graph output from the improved model,and verified on the test set.Experimental results show that the accuracy of the pose recognition model is 92.5%,which is suitable for the application of mobile robots.In this paper,the improved object detection model,pose estimation model and pose recognition model are applied to the robot based on ROS system,and an object detection and pose recognition system is designed.The experimental results show that the real-time performance and accuracy of the system are good,and the system can reach 25 FPS.It can also work well under some occlusion conditions.