基于里程信息融合的株间锄草刀定位数据优化方法

为提高株间锄草刀定位精度、降低机器视觉受外界因素的影响，该文提出里程信息融合机器视觉的方法对锄刀定位数据进行优化。通过分析定位数据校正和视觉滞后补偿的原理，设计了模糊逻辑校正器，通过模糊规则将模糊校正系统简化为单输入单输出形式，采用Mamdani模糊推理方法获得视觉数据可信度决策表，将可信度作为加权值生成校正锄刀定位数据，并提出采用实时里程信息作为视觉滞后补偿量的方法，给出补偿公式。田间刀苗距优化静态试验表明，视觉刀苗距误差为9.88 mm，优化后刀苗距误差为6.06 mm；动态试验表明，视觉数据出错率为4.8%～6.6%，刀苗距变化曲线显示，优化方法可有效过滤视觉坏点或不稳定的数据点，将视觉滞后纳入衡量标准，不同车速下动态优化后刀苗距平均误差为5.30～7.08 mm，较优化前降低了25%左右。研究结果表明，锄草刀定位数据优化方法可有效提高机器视觉静态和动态获取刀苗距的精度。该研究为提高株间锄草技术的锄刀定位精度提供了参考。

Optimization method of intra-row weeding hoe positioning data based on odometry information fusion

Abstract: The accurate positioning data of intra-row weeding hoe can provide the basis for intelligent intra-row weeding robot and influence the effects of weeding directly. The main method of crop and weed information acquisition is based on computer vision which has excellent real-time performance, high accuracy, low cost and other benefits. But light intensity, appearance of the crop, shadow, missing plant, weed density, mechanical vibration or other conditions could degrade the performance of machine vision. This work describes an optimization method which includes the correction algorithm and the visual lag compensation algorithm based on the fusion of odometry and computer vision for improving the accuracy of intra-row hoe positioning data. In this work, the fuzzy corrector was designed for fusing odometry data and vision data. Fuzzy correction system was simplified as a form of single input and single output by the fuzzy rules achieved earlier. The Mamdani fuzzy inference method was used to obtain the reliability and weighted value of vision data, and then, a new corrected positioning data could be created by weighted values of 2 sensors. Because of the time-consuming problem in image processing, the hoe positioning data received by processor and the actual hoe positioning data were not equal. Using odometry information which could be calculated by the pulse signal of rotary encoder as a compensation of visual delay was proposed and the formula for calculating compensation was given. To assess the performance of optimization method, 2 sets of field experiments which consisted of static and dynamic tests were designed for detecting the correction accuracy and the compensation precision. The hoe positioning optimization system was equipped with the weeding robot which was connected with tractor by the front three point linkage system. In the static trails the tractor was randomly stopped, and the vision data and optimized data of hoe position were received by processor; then the distance from hoe to crop was measured as a standard for calculating the error of 2 sets of data. Experimental results showed that the average errors for vision data and optimized data were 9.88 and 6.06 mm respectively, and the error after optimization decreased compared to that before optimization. In dynamic experiment, hoe positioning data were collected in real time and the curve of data change was drawn. Curve analysis showed that the error rate of vision data was 4.8%-6.6% and optimization method could effectively filter the error and unstable vision data points. The average error of optimized data was 5.30-7.08 mm at different speeds, about 25% less than before. Research results showed that the error data would occur in intra-row weeding system based on machine vision in the field environment. The fuzzy correction algorithm and the visual lag compensation algorithm could effectively judge and filter the wrong visual data points and improve the accuracy of hoe positioning data and the stability of system under static and dynamic conditions. The method mentioned in this paper can provide the theoretical basis for precise hoe positioning of intra-row weeding technique and the technical reference for the related researches on weeding robot.