Simultaneous localisation and mapping in a complex field environment

The usefulness of image registration techniques in mapping and localising a robot in an agricultural environment by using readings from a laser range scanner was investigated. The proposed approach used frequency domain and correlation. Translational and rotational differences that occur between successive readings of the scanner and that correspond to the movement of the robot were used. The approach was tested on 9 test runs, with a total of 252 m in length, recorded in an apple orchard and in a vineyard. The results were then compared to results from the Hector mapping algorithm. It was shown that the present approach performed very well compared to Hector mapping. On average achieved an 4.24% ± 2.9% error rate and the present approach 0.16% ± 0.1%. Hector mapping on the other hand proved better in cases where rotational differences were looked for, reaching an error rate of 1.69% ± 0.7% in comparison to present approach with an error rate of 4.19% ± 3.1%. •We are focussing on a autonomous agriculture machines applications.•Proposed an new approach for complex crop field mapping.•Advantage of cheap and simple backbone for autonomous agriculture tasks.•Algorithm have been compared to other state-of-art mapping algorithms.