Research On Visual Recognition And Local Path Tracking Of Road Environment For Horticultural Electric Tractor

Modern agriculture is developing towards diversification,in which intelligent tractors are becoming a hot research topic.The intelligent tractor autonomous walking system is mainly composed of three subsystems,namely,environmental awareness system,decision control system and execution mechanism system.This paper combines the key research and development plan(modern agriculture)key projects in Jiangsu Province: Horticultural Electric Tractor Research and Development(BE2017333),to study the autonomous walking system of horticultural electric tractors,the main work is as follows:(1)According to the needs of the horticultural electric tractor autonomous walking system,on the basis of the original tractor,from the environmental perception,decisionmaking control and implementation of the three aspects,select the appropriate hardware,and build the hardware-based up and down machine software system,and finally through debugging to complete the construction of the horticultural electric tractor autonomous walking system,to provide a test platform for the follow-up research.(2)Using the camera to collect orchard environment photos on the spot,preprocess the photos,build orchard environment data sets through artificial labeling and data enhancement,design orchard environment recognition models,and develop local path tracking navigation lines.In view of the complex characteristics of orchard environment,an orchard environment recognition model based on deep residual U-type network is proposed,which adds residual learning in the coding layer,combines jump connection and residual block,improves the model depth and information fusion ability,and proves that the model has high recognition accuracy and strong robustness through data set training and verification.On this basis,the driving road recognized by the depth residual U-type network model is extracted by The Canny Edge Detection Algorithm,the central line idea and the minimum two-way method are used to fit to generate the tractor local path tracking navigation line,and finally the pixel coordinates of the navigation points in the navigation line are converted to the world coordinates to meet the actual use needs.(3)Study the local path tracking control algorithm of horticultural electric tractors.Based on the tractor’s two-degree-of-freedom model and the single-point pre-view trajectory prediction model,the steering wheel corner required by the tractor from the current position to the pre-view point position is calculated,and the driver’s manipulation model is tracked by the incremental square single-point pre-view trajectory.In view of the problem that fixed forward distance affects tractor track tracking ability,an adaptive forward distance fuzzy controller is designed to further optimize the local path tracking control algorithm of tractor.Through the joint simulation of CarSim and Simulink,the effectiveness of local path tracking control algorithm of horticultural electric tractors is verified.(4)Experimental verification.The road recognition algorithm and local path tracking control algorithm were transplanted into the autonomous walking system of horticultural electric tractor,the suitable test site was selected,the performance of the visual-based garden tractor autonomous walking system was tested,the test results were analyzed,and the suitability and stability of the whole system were verified.