Shufeng Han

Obstacle detection using stereo vision to enhance safety of autonomous machines

As semi-autonomous and autonomous vehicles are finding more and more applications in agriculture, the capability of detecting obstacles in the field and taking appropriate safety actions is becoming more critical. This article presents an obstacle detection system using a binocular stereo camera. Field tests using a standing person as the potential obstacle indicated that the system could reliably detect the obstacle and its motion status in the vicinity of an agricultural vehicle, although the accuracy of the system decreased as the range between the vehicle and obstacle increased.

Kalman Filter for Sensor Fusion of GPS and Machine Vision

Field experience with agricultural vehicles has shown the benefit of merging local and global positioning information for a satisfactory performance of autonomously guided vehicles. Global localization sensors provide an adequate procedure to determine the absolute position of a vehicle, but lack the accuracy of local positioning sensors to make available up-to-date centimeterrange spatial information. This project combines the global positioning data obtained from a DGPS receiver with the local information generated by a monocular camera mounted on the front of an agricultural tractor. The fusion technique applied to blend the data from both sensors is based on the Kalman Filter. The selection and definition of the model states and measurements, as well as the noise covariance matrices turned out to be key points in the system behavior. The algorithm was implemented on a tractor driven along an artificially-created curve. Simulation and field test results are presented and discussed in the manuscript. The Kalman filter was able to reduce the effects of noisy trajectories resulting in more stable control commands.

Fuzzy Logic Model for Sensor Fusion of Machine Vision and GPS in Autonomous Navigation

This article presents a framework for characterizing GPS errors typically found in agricultural applications of autonomous vehicle guidance. A fuzzy logic model was proposed to combine sensor data from a differential GPS receiver and a machine vision perception engine. The objective was to correct GPS tracking errors with local positioning information provided by a monocular camera. The fusion philosophy was based on estimating the quality of each sensor output and obtaining the positioning corrections according to a quality-based weight. Field experiments demonstrated the adequacy of using a local positioning sensor to correct a global positioning receiver’s error through sensor fusion. The fuzzy logic model was implemented to guide a tractor successfully at various speeds and under diverse field and sensor signal quality conditions.

Evaluation of automatically steered agricultural vehicles

The popularization of automatic guidance systems for off-road vehicles, especially in agricultural environments, has led to a sudden affluence of commercial solutions. As a consequence, end users are often overwhelmed by the spread of possibilities commercially available, on top of the confusion provoked by the guidance accuracy advertised by manufacturers. Therefore, there is a need for a methodology that objectively quantifies the level of performance of any auto-steering system, classifying different solutions according to their accuracy and reliability. This paper describes a method to evaluate the behavior of any automatically driven agricultural vehicle traveling along paths of any curvature. The proposed technique was verified in a self-propelled forage harvester, and field results are presented and compared with those found when the harvester was evaluated with conventional regression analysis.

Trajectory-Based Algorithm for the Quantitative Evaluation of Automatically Steered Vehicles

During the last ten years, automatic guidance systems have become more common in agricultural vehicles. However, users of auto-guided systems can be confused by the growing variety of options commercially available, as well as by the guidance accuracies advertised by different manufacturers. This work proposes an algorithm to evaluate the performance of auto-steered machines for any kind of vehicle and any type of guidance system in the general case of straight row and curved row guidance. The core of the algorithm is based on the comparison of two trajectories: reference course and actual path. The algorithm searches in a neighboring area for the reference points and calculates the deviations. Statistical analysis of the errors provides quantitative information to evaluate the behavior of auto-steered vehicles. The advantage of this technique rests on its independence of regression methods and its immunity to outliers. This methodology was applied to an automatically guided self-propelled forage harvester at both low-speed and high-speed guidance. Results are presented, and when compared to those obtained applying conventional linear regression, they show a slighter impact of outliers and a more efficient procedure and data management.

Obstacle Detection Using Stereo Vision To Enhance Safety of Autonomous Machines

As semi-autonomous and autonomous vehicles are finding more and more applications in agriculture, the capability of detecting obstacles in the field and taking appropriate safety actions will become more critical. This paper presents an obstacle detection system using a binocular stereovision camera. Field tests using a standing person as the potential obstacle indicated that the system could reliably detect the obstacle and its motion status in the vicinity of an agricultural vehicle, although the accuracy of the system decreased as the range between the vehicle and obstacle increased.