A lightweight and fast target detection method for hand-eye manipulator system is presented

Abstract

A lightweight and fast target detection method for a hand-eye manipulator system relates to thefield of computer vision, The invention particularly relates to a lightweight and fast target detection method applied to a hand-eye manipulator system, which comprises the following steps: first, collecting a training set: collecting pictures under various conditions according to the targets in actual application scenes, and fully collecting pictures of targets under various background conditions under various illumination conditions; 2, designing a depth convolution network model and training; Step3: Applying the trained model to the object detection in the actual environment, and output the category and position of the object in the picture. As that technical proposal is adopted, the invention has the beneficial effects that the target detection model can not only guarantee the actual use precision performance requirement, but also meet the real-time requirement, reduce the storage cost and calculation cost of the original method on the embedded platform, and improve the work efficiency of the service object of the target detection system.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a lightweight and fast object detection method applied to a hand-eye manipulator system. Background technique [0002] In recent years, with the development of computer technology and the improvement of computer hardware performance, the application performance in the field of computer vision has been greatly improved. Target detection, as an important and basic task, has significantly improved its recognition accuracy and positioning accuracy. [0003] The traditional target detection method mainly uses the method of HOG feature + SVM classifier. However, HOG features are artificially designed, which cannot meet the performance requirements of object detection with arbitrary poses in natural images. With the development of deep learning technology, it has achieved great success in the field of target detection, and its performance has completely surpassed traditional detection m...

AI Technical Summary

Problems solved by technology

But even for the original version of the YOLO series, its real-time performance can only be guaranteed on top graphics cards (such as titan x)

If it is transplanted to an embedded platform (such as TX2) without processing, its real-time performance cannot be guaranteed

[0005] The network model trained by the original method is relatively large, so that the real-time requirements cannot be guaranteed when the model is transplanted to the embedded platform (TX2), which seriously affects the work efficiency of the target detection system service object (such as the hand-eye robotic arm system)

Images