Unmanned aerial vehicle (drone) uses a number of functional to perform assigned flight tasks. One of these systems is the navigation system. Usually for navigation during flight it uses a GPS, but

Unmanned aerial vehicle (drone) uses a number of functional to perform assigned flight tasks. One of these systems is the navigation system. Usually for navigation during flight it uses a GPS, but it does not always ensure absolute reliability and correct operation due to the fact that the basis for the operation is in constant contact with the system of satellites that can leave the visibility area. To support the use of navigation systems other orientation systems are used, such as navigation by terrain and landmark navigation.

This article offers the algorithm of the landmark navigation system. For the given algorithm operation the satellite images of the area with reference to the actual GPS coordinates are loaded to the on-board computer of UAV. On these pictures reference objects that stand out clearly against the background areas are selected: free-standing structure, unique twists of roads or rivers, in flight over the city it can be any building that can be easily identified against the general background and is proximately close to the actual route. Then during the flight in case of loss of GPS satellite signal or using the system of reference points as a basic, with the help of a special video recording equipment flight inspection of the area, recognition of objects passing between them and the necessary marks search is done. In case it is found, onboard system confirms your current location and continues to search for the next mark in line.

Operation and algorithm

The fundamental mechanism of the landmark navigation system is the contour analysis system. It is expected that the contour contains the information about the object shape. The internal points of the object are not taken into account. This limits the scope of contouring analysis algorithms, but considering only contours you can move from a two-dimensional image space to the space of contours and thereby reduce the computer and algorithmic complexity.

KA (не знаю что это) can effectively solve the problems of pattern recognition - transposition, rotation and scaling an object.

This option is provided by the use of autocorelative functions that are constant or close to the value for the objects in the states mentioned above.

Firstly we define the contour of the object. Contour is the limits of the object, an array of points (pixels) that separate the object from the background.

In machine vision systems, some used contour encoding formats are Freeman code, two-dimensional coding. But all these encoding formats are not used in KA(опять).
Instead, the KA(що це,в дідька, таке?) contour is encoded by a sequence of complex numbers. A point on the contour is selected, it is used as a starting point and is fixed. Then the contour is scanned (say - clockwise) and each displacement vector is observed on complex plane a + IB. Where a is displacement on the X axis and B - on the y-axis. The shift is marked relatively previous point.

Due to the physical nature of three-dimensional objects, their contours must always be closed - it eliminates self-crossing. This allows you to clearly define how the bypass of a contour is done (up to a direction - clockwise or counter -clockwise). Last vector path always leads to the starting point.

Each vector- contour will be called basic vector (BV). And the sequence of complex numbers - vector contour(VC).

Vectors – contours will be denoted by large Greek letters and their elementary vector - small Greek letters.

Thus, the vector - contour Γ of length to can be described as:

Contours, as vectors of complex numbers possess excellent mathematical properties, as compared to other modes of coding.

In general, the complexity of the encoding is close to two-dimensional coding, where the contour is defined as the set of points represented in two-dimensional coordinates. But there is a difference between operation with the scalar product of vectors and complex numbers. This is what gives priority to the KA methods.

Contour Properties

• The sum of closed contour vectors is zero.

• Contour vector does not depend on the parallel transfer of the original image.

• Image rotation angle is equivalent to a rotation of each elementary vector contour at the same angle.

• Modification of starting point leads to a cycle shift. As elementary vectors are encoded relatively the previous point, it is clear that the modification of the starting point, the sequence of elementary vectors will be the same, but the first elementary vector will be the one that starts with the starting point

• Scale source image can be seen as multiplying each elementary vector contour to a scale factor.

The properties of the autocorrelation functions

• Is not dependent on the choice of the starting point of the contour

• As АКФ is the sum of pairwise works of EB contour each pair occurs twice in the range from 0 to K.

• If the circuit has certain points of symmetry, АКФ will also have this symmetry

• Normalized АКФ is not independent on the scale, position, rotation, and starting contour point selection.

Using the above properties, it is possible to implement an adaptive flight conditions UAVs extraction of predefined object contours algorithm which, by АКФ can only be proximate but can not repeat exactly those that are used to program the navigation system.

An example of the algorithm is given in Annex

Appendix 1. Search for and recognition of the national Shevchenko park.

Appendix 2. Distinguishing bridge in Kyiv near metro station "Shuliavka"

Дата добавления: 2015-11-04