Research On Scene Analysis And Path Planning Algorithm Of Picking Robot

With the development of emerging agricultural production models and technologies,agricultural robots will become the backbone of agricultural production.As an important component of agricultural robots,picking robots are of great significance for accelerating the process of agricultural modernization and reducing labor costs.In a complex orchard environment,the autonomous movement of the picking robot is the technical basis for the fruit picking task.The scene parsing and path planning of the picking robot integrated with multiple technologies have become a research hotspot.To improve the scene parsing and path planning capabilities of the picking robot on the orchard road,this paper studies the scene parsing and path planning algorithm of the picking robot based on the design and construction of the picking robot.The main research contents are as follows:In this paper,the overall design of the picking robot for the orchard environment.The picking robot adapts to the terrain of the orchard road and the ridge planting area through the passive omnidirectional cross-country movement unit.It uses binocular cameras for realtime environmental perception and uses a global satellite navigation module and attitude module to obtain position and attitude information.It installs a horizontally movable manipulator for fruit picking and uses an onboard PC and underlying controller for information processing.Through the integration of various modules,this paper has completed the construction of the picking robot required for the research.Aiming at the difficulties of feature extraction and poor real-time performance for picking robots in the orchard road environment,a multi-branch pyramid scene parsing network(Mu-PSPNet)is proposed.The consists of Mu-PSPNet is an encoder and decoder.The encoder,which uses PSPNet for feature map extraction,applies the pyramid pooling module to obtain rich multi-scale features.It reduces network parameters and computing time by introducing deep separable convolution.The decoder exploits a multi-branch attention module for multi-channel feature fusion and uses a global context module to supplement semantic information to improve the network’s ability to acquire contextual semantic information.The experimental results show that the Mu-PSPNet model obtains75.6% and 73.4% MIo U on the self-built orchard road dataset and the SSSDC public dataset respectively.It reaches a scene parsing speed of 10 FPS,which improves the real-time performance of the scene parsing for agricultural intelligent robots.To solve the obstacle detection problem of the picking robot in the orchard road environment,the binocular stereo vision obstacle detection method fused with semantic information is realized.The Zhang Zhengyou calibration method and Bouguet algorithm are used to calibrate the binocular camera and the epipolar line respectively.Moreover,the semiglobal block matching algorithm is used to achieve binocular stereo matching to obtain parallax images.The obstacle type and distance information are obtained under which the contour information and the center of gravity coordinates of different obstacles are obtained by processing the semantic label map image.By fusing the semantic information and the disparity map,the obstacle type and the distance information of the obstacle are determined.The experimental results show that the average error of obstacle distance measurement is2.52%,which meets the requirements of road obstacle detection for picking robot orchard.Aiming at the path planning of the picking robot in the orchard road environment,a dynamic window local path planning algorithm based on the road guide line is proposed.In this paper,the discrete guide points are obtained by processing the semantic label map of the orchard road area,and the B-spline curve is used to fit the discrete guide points to obtain the road guide line.The guide line information is introduced into the dynamic window method through the evaluation function for local path planning.Carrying out the guide line accuracy test,the result shows that the road guide line is extracted accurately,and the relative error is2.25% on average.The local path planning experiment of the picking robot is carried out,and the results show that the dynamic window method based on the road guideline has good guideline following and obstacle avoidance capabilities.