Research On Generating Guidance Line Algorithms In Agriculture And Forestry Environments

The vision guidance technologies of an autonomous robot for agriculture and forestry usage are studied in this paper. Natural scenes are classified to help identify corresponding paths embraced by vegetations. Path information is then abstracted from its background using reliable and steady image processing methods, and the central line of path is figured. Algorithms of guidance line detection based on Hough transform are proposed and advanced to provide quantified real-time information of the path, the angle and horizontal offset according to the image vertical central line. CCD is equipped as the main image collecting device with monotonous vision pattern.A few path guidance line generation algorithms are proposed and discussed in several natural scenes. MT-R intelligent robot is introduced and programmed with the basic algorithm to validate under simulated natural scenarios. Main researches are:1. Basic image enhancement techniques are analyzed and applied to the images taken in these environments. Noises are added to the images and their effects are compared. Denoising algorithms are compiled to weaken the effects. Spatial filtering and smoothing methods are used.2. Image segmentation methods are studied upon different scene characteristics. An algorithm is proposed to obtain steady image segmentation results in different natural-scene images.3. Guidance line abstraction method is proposed based on the idea to distill both boundaries of the path and then calculate the line through the path center. In addition, different natural scenes are discussed to make adjustment to the algorithms and value setting. Simulation results show that the image segmentation is steady and in line with human vision judgments.4. Guidance line generation methods are proposed and applied to the images. Based on Hough transform, the peak detection process is modified to address complicated natural environments: one is to pre-limit the angular value of possible collinear points before peak detection. This enables swift key-info abstraction in orderly-planted fields while offsetting defects caused by previous image segmentation; the solution to generate guidance line in forestry environments and the alike is to locate points of trunk-and-earth intersection, then a point array in middle of the two point series is obtained. Next, two methods are applied to process the dot array: either to detect a line for guidance using Hough transform, or to least-square fit the two points series so a line through the center is acquired as the guidance path.5. Algorithm proposed above is also applied to fields with tall crops similar to trees in a forest.6. Camera model is set up and studied to introduce a wildly used calibration method. Itmakes the vision system easier to use and need not exact and costly equipment for calibration.8. The basic method to trace a curve on the ground is programmed and performed on MT-R,the intelligent robot. Results show efficiency of the algorithm.