Robot weak light environment grabbing detection method based on multi-task sharing network

Abstract

The invention discloses a robot weak light environment grabbing detection method based on a multi-task sharing network, and belongs to the technical field of computer vision and intelligent robots. According to the method, strict matching of shot images in a weak light environment and a normal light environment is ensured through image acquisition, a corresponding data set d is constructed, then Darknet-53 is adopted as a backbone network to construct a weak light environment capture detection model, multi-scale features with strong feature expression ability are extracted, and, through a parallel cascade capture detection module and an image enhancement module, capturing detection and weak light image enhancement tasks are realized respectively; training samples are randomly selected from the data set d, a weak light environment capture detection model is used for prediction, and when the change of a loss value within the iteration number iter is smaller than a threshold value t, the capture detection model G is converged, that is, training of the model G is completed; and an image Ilow shot in the weak light environment is input into the trained model G to obtain a predicted capture frame parameter and an enhanced image, and a capture detection task in the weak light environment is completed.

Description

technical field [0001] The invention belongs to the technical field of computer vision and intelligent robots, and in particular relates to a method for grabbing and detecting robots in weak light environments based on a multi-task sharing network. Background technique [0002] With the continuous development of artificial intelligence, robot control and perception technology, robots equipped with multi-degree-of-freedom robotic arms can autonomously realize the grasping operation of different target objects, and carry out advanced human-computer interaction. Fields are playing an increasingly important role. A typical robot grasping process generally includes four steps: target positioning, pose estimation, grasping detection, and motion planning. Among them, grasping detection can determine the graspable parts of the object, which directly determines the stability and stability of the robot's grasping operation. Accuracy plays an important role. At present, the robot gra...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem of low detection accuracy in low-contrast and low-light environments, the present invention provides a robot-based low-light environment grasping and detection method based on a multi-task sharing network

Images