DE102020127964B4 - Method and apparatus for rapidly capturing a hyperspectral cube of an object or scene - Google Patents

Abstract

A method of determining a hyperspectral (HS) cube of an object or scene, comprising the steps of:a. Providing a mobile electronic camera with a static, spatio-spectral camera attachment and at least one inertial sensor rigidly connected to the camera;b. Moving the mobile camera along a trajectory and/or rotating the camera about at least one camera axis such that the object or the scene remains at least partially in the field of view of the camera, whereinc. continuously time-indexed spatio-spectral images and inertial sensor readings are captured by the camera during movement and/or twisting;d. determining displacement transformations from pairs of time-indexed images using the relative motion and rotation of the camera determined for each pair of images from the time-indexed inertial sensor readings;e. Reconstructing the HS cube by populating a three-dimensional data array with measured intensity values from the spatio-spectral images using the determined shift transformations.

Description

The invention relates to a method for spatially resolved, electronic detection of the electromagnetic radiation emanating from an object or a scene in a spectral range that can include ultraviolet, visible and infrared wavelengths. The invention further relates to a camera attachment for a conventional electronic camera that enables the method to be carried out.

The color photograph of any object (or of a scene, this is not often mentioned anymore for the sake of brevity) is known to the general public as a tried and tested means of conveying quite complex pictorial information to the viewer. In addition to its position in the image - the image coordinates - each pixel is also assigned color information, namely, for example, a classification as red, green or blue (RGB color space) or a mixture of these colors. Humans translate wavelength ranges of the VIS spectral range into colors, but for a machine, especially a computer, the color photograph is nothing more than a three-dimensional data set of intensity values associated with image coordinates and wavelengths. The computing machine receives these intensity values as measured values from an electronic camera, with the camera also supplying the 3D coordinates.

The "sight" of electronic cameras surpasses the human eye in the width of the detectable spectrum. The light-sensitive pixels of the camera chip react not only to visible (VIS) light, but also to ultraviolet (UV) and infrared (IR) light, thereby allowing non-contact and non-destructive insights into an object's physical and chemical properties.

Furthermore, if the camera is configured to capture a continuum of wavelengths, then it can be used to determine a hyperspectral (HS) cube of an object. An HS cube is a measured 3D dataset about an object, just like a photograph, from the camera's perspective of the object, from - i. i.e. R. adjustable - camera parameters and depends on the lighting of the object. As an inevitably finite data set, it is limited in its resolution, just like a photo, both in terms of image coordinates and wavelengths. The HS cube is an image of the real light emission of an object, which can be fed to a computing unit, usually a microprocessor, for storage and further processing.

The term HS cube is common in the literature and should not be understood here as a limiting geometric cube. In particular, the two image coordinates can also run over length intervals of different lengths. Even a circular image coordinate area, which would lead to an HS cylinder after viewing, should be referred to as an HS cube in the context of this description. Furthermore, the boundaries of the detected wavelengths can vary across the range of image coordinates even when dealing with an HS cube.

A numerically recorded HS cube is a database and a necessary prerequisite for a large number of interpretive applications. For example, the work of Zhao et al., 2009, "Nondestructive measurement of sugar content of apple using hyperspectral imaging technique", Maejo International Journal of Science and Technology Vol. 3, pp. 130-142 deals with the study of the shelf life of fruits and determining their degree of ripeness. Further work examines the pesticide load on plants, the development of bacterial cultures in food and the putrefaction process in meat products. Medical applications lie in the non-invasive diagnosis of tumor cells, e.g. from Fei et al., 2017, Label-free reflectance hyperspectral imaging for tumor margin assessment: a pilot study on surgical specimens of cancer patient, J. of Biomedical Optics, 22( 8), 086009 (2017).

Capturing an HS cube harbors the inherent problem of measuring an intensity function, which is explained for three independent coordinate axes - here: image coordinates x, y and wavelength λ - with a 2D detector - the camera chip. The usual solution consists in capturing a plurality of parallel cutting surfaces (usually cutting planes) through the 3D data set and subsequently supplementing the portion of the 3D data actually captured in this way with estimated or interpolated data.

For example from the publication US4455087A is an apparatus that filters the polychromatic light coming from an object using an adjustable double-grating monochromator in such a way that a monochromatic image of the object can be captured. The monochromatic image is a sectional plane through an HS cube at a fixed, predetermined wavelength. The wavelength that is to be captured by the camera can be selected by synchronously tilting the diffraction gratings in the device. In the course of a scan across the wavelengths, a predetermined number of images can be acquired, which when viewed together is suitable for the reconstruction of an HS cube.

Among other things, from the work of Meng et al. "Double-pass grating imaging spectrometer", CHINESE OPTICS LETTERS, 17(1), 011202(2019), an optical structure is known that allows the detection of a sectional plane running diagonally through an HS cube in a camera image. One of the two image coordinates is optically linked to the wavelength, so that an image of an object is captured in which the wavelength varies along one of the image axes. The link is made by Meng et al. through a transmitting diffraction grating, to which the collimated polychromatic light emanating from each object point is fed. The grating spectrally disperses the light and distributes it to a continuum of exit angles along its direction of dispersion. The light is then focused at one of the exit angles onto a slit mirror and collimated onto the grating. However, only that part of the object light arrives at the mirror for which the angle of incidence on the grating and the wavelength fulfill exactly those diffraction conditions that lead to the light exiting the grating in the direction of the mirror. It is precisely this light that is redistributed to its original entry angles on the second pass through the grating and imaged on the camera. If the dispersion direction of the grating is oriented along an image coordinate, this entails a linear wavelength distribution on this very image coordinate.

A diagonal section through an HS cube is obtained directly as an image. Such an image is also called a spatio-spectral image, or sometimes more intuitively, a rainbow color image. Meng et al. show sample images in which scenes with wavelengths that vary along an image coordinate in the VIS spectrum - in rainbow colors - can be seen.

In Meng et al. is intended to determine an HS cube by changing the assignment of wavelength and image coordinates by pivoting the mirror in order to detect other exit angles behind the grating. In this way, a sequence of spatio-spectral images can be generated, with each pixel in each image showing the emission of a different wavelength. An HS cube can then be reconstructed from the synopsis of the image sequence.

However, it should be noted that a spatio-spectral camera can also be a monochrome camera in which the camera pixels themselves cannot distinguish colors. Monochrome cameras usually have a higher resolution than color cameras. The wavelength information is nevertheless available, since the imaging parameters of the lens (in Meng et al.: Telescope), the grating constant and the exit angle at which the mirror is arranged are known. In addition, a reference object that emits one or more defined wavelengths can be arranged on the object. In the course of a scan, the reference object appears particularly bright if its position in the image and its previously known emission just match the spatio-spectral coding of the camera system. Using one or more such known fixed points in the HS cube, its numerical reconstruction can also be accomplished from monochrome image data and the wavelengths can be subsequently determined.

The work of Kim et al., "Smartphone-based multispectral imaging: system development and potential for mobile skin diagnosis", Biomed. Opt. Express 7, 5294-5307 (2016) points out that there is a lot of interest today in new technologies for telemedicine and/or healthcare at home. She suggests converting a conventional smartphone so that it can take meaningful pictures of living skin for the early detection of skin diseases. The method of choice for the authors is to acquire a multispectral cube of the skin in a local area of interest, e.g. at an acne pimple. They use the LED integrated in the smartphone for lighting, the spectrum of which is already known, and a filter wheel with nine narrow-band color filters. The color-filtered light is applied to the skin in a fanned-out pattern, with reflected light that does not penetrate the skin being blocked out by crossed polarizers. The light scattered by the skin is detected, with one image being recorded for each of the color filters by turning the filter wheel further between the recordings. Due to the limited and discontinuous range of wavelengths that can be detected, one speaks of a multispectral cube instead of a hyperspectral one, for which the wavelengths are continuously scanned over an interval.

Kim et al. have designed an attachment for a standard smartphone, with which the usual system of integrated camera chip and integrated LED of the smartphone can be upgraded to a multispectral measurement system. This is a good idea to make multi- or hyperspectral image acquisition more widespread and thus new potential applications, which is currently still being hampered by the high acquisition costs for commercial expert systems.

Furthermore, Wilcox CC, Montes M., Yetzbacher M., Edelberg J., Schlupf J., "An ultracompact hyperspectral imaging system for use with an unmanned aerial vehicle with smartphone sensor technology", Proc. SPIE 10639, Micro- and Nanotechnology Sensors, Systems, and Applications X, 1063919 (8 May 2018); https://doi.org/10.1117/12.2303914 used a compact hyperspectral image sensor with multiple Fabry-Perot etalon arrays. Using this sensor it is possible to scan the ground line by line in a series of pixels, creating a hyperspectral data cube and georeferencing to a digital surface model of the ground with GPS INS parameters for location (latitude, longitude, altitude) and attitude (azimuth, pitch, roll) is done with 6 degrees of freedom.

U.S. 2017/0264833 A1 describes a method for acquiring and consolidating multispectral image data for specific regions using one or more spectral enhancement mechanisms. In a first multispectral or hyperspectral workflow, time-stamped two-dimensional pixel images of a specific region are generated with the aid of a computer-assisted imaging device. Each of the images is based on image data acquired for a specific frequency band from a plurality of frequency bands.

Further disclosed US 2021/0093257 A1 a method and system for diagnosing a medical condition of a target area of a patient using a mobile device. One or more magnetic field images and hyperspectral images of the patient's target area are generated and a spatial position of the mobile device relative to the patient's target area is recorded for each of the images. Based on the images and data obtained, a 3D image of the target area is generated, which can be used for diagnostics or monitoring of the patient U.S. 6,166,373 A discloses a focal plane scanner having a front objective lens, a spatial window for selectively transmitting a portion of an image, and a CCD array for receiving the transmitted portion of the image. It is essential for the technical solution described that the spatial window and the CCD array are designed for simultaneous movement relative to the front objective lens and that the spatial window is arranged within the focal plane of the front objective.

the CN 105865624A discloses a spectral extraction method and apparatus for a hyperspectral collection system, the hyperspectral collection system including a slit, a smartphone, a light splitting module, and a collimation module.

However, all of the above systems have the feature that in order to carry out a wavelength scan, they have to make changes to the optical structure in front of the camera chip, typically the precise rotation or pivoting of at least one optical component using suitable electrical drives, e.g. stepper motors. In addition, the drives must be synchronized with the camera shutter so that the settings of the optical components are well-defined and known for each image capture. A standard smartphone usually does not have an interface for controlling such drives (Kim et al. built one) and also does not provide a power supply for it. A static camera attachment that can do without moving components, drives and controls would therefore be desirable.

The article by Sima et al., COMPACT HYPERSPECTRAL IMAGING SYSTEM (COSI) FOR SMALL REMOTELY PILOTED AIRCRAFT SYSTEMS (RPAS) - SYSTEM OVERVIEW AND FIRST PERFORMANCE EVALUATION RESULTS, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI- B1, 2016, XXIII ISPRS Congress, 12-19 July 2016, Prague, presents a camera system that has a camera chip with a linearly variable interference filter fixed on the camera pixels. In this way, the pixels capture different wavelengths in columns, i.e. here, too, rainbow-colored images are created. The camera is mounted on a flying drone looking down and captures an image sequence with 72 wavelength bands of light in the interval 600 - 900 nm for each object point flown over during the flight. The wavelength scan is thus carried out by moving the camera along the object or the scene, with correspondences between the images in the sequence being determined and used in order to allocate the spectral intensities to locations in the images. Such correspondences can be determined by landmarks in the images, e.g. prominent edges (“features”), or by area similarities, e.g. two-dimensional cross-correlation between images.

In the case of the landmark-based algorithms, the intensity edges in the respective spectral ranges are used to find the location coordinates of the respective landmark. The image coordinates of the same landmarks in two different images are assigned to one another and displacement transformations are calculated therefrom. In the simplest case, this is a simple vector plus rotation and, if necessary, scaling, but it can also be a vector field with the displacement vectors of the distortion between the two images under consideration. Algorithms that use cross-correlation also exploit intensity variations across the image area without considering landmarks individually. Information about the displacement, rotation and, if applicable, scaling is thus also obtained, collectively referred to here as Displacement transformations between pairs of images.

For the two fundamentally different evaluation options, it is assumed that spectrally adjacent image information has similar intensity variations such as edges, for example, and that the two images under consideration do not have too great a mutual shift. If the image scenery contains too few intensity variations or if these are unfavorably distributed in the spatial frequency spectrum and have too little contrast, adding an additional high-contrast reference object, for example with strong black-and-white contrast in all spectral bands and non-repeating pattern for both evaluation options, can provide sufficient information in the image area for Will be provided.

Although the Sima et al. with the linearly variable filter on the camera pixels, is a compact and convenient design for capturing spatio-spectral images, but it has disadvantages. On the one hand, it is a special construction that cannot be created from a conventional camera in a few simple steps - e.g. with a camera attachment - but requires a careful coating of the camera chip. The conversion cannot be easily reversed either, i.e. it can no longer take normal pictures. In addition, the wavelengths that each individual pixel can detect are each finally fixed to a predetermined, narrow interval.

The object of the invention is to formulate a method for capturing a hyperspectral cube of an object that provides for the use of a commercially available electronic camera with a static, spatio-spectral camera attachment.

1. a. Bereitstellen einer mobilen elektronischen Kamera mit einem statischen, spatio-spektralen Kameraaufsatz und wenigstens einem starr mit der Kamera verbundenen Inertialsensor;
2. b. Bewegen der mobilen Kamera entlang einer Trajektorie und/oder Verdrehen der Kamera um wenigstens eine Kameraachse, derart dass das Objekt oder die Szene wenigstens teilweise im Sichtfeld der Kamera verbleibt, wobei
3. c. während des Bewegens und/oder Verdrehens fortlaufend spatio-spektrale Bilder und Inertialsensormesswerte von der Kamera zeitlich indiziert erfasst werden;
4. d. Ermitteln von Verschiebungstransformationen aus Paaren von zeitlich indizierten Bildern unter Verwenden der für jedes Paar von Bildern aus den zeitlich indizierten Inertialsensormesswerten ermittelten relativen Bewegung und Verdrehung der Kamera;
5. e. Rekonstruieren des HS-Kubus durch Belegen eines dreidimensionalen Datenarrays mit gemessenen Intensitätswerten aus den spatio-spektralen Bildern unter Verwenden der ermittelten Verschiebungstransformationen.

The object is solved by a method for determining a hyperspectral cube of an object or a scene with the steps:

1. a. Providing a mobile electronic camera with a static, spatio-spectral camera attachment and at least one inertial sensor rigidly connected to the camera;
2. b. Moving the mobile camera along a trajectory and/or rotating the camera about at least one camera axis such that the object or the scene remains at least partially in the camera's field of view, wherein
3. c. continuously time-indexed spatio-spectral images and inertial sensor readings are captured by the camera while moving and/or twisting;
4. i.e. determining displacement transformations from pairs of time-indexed images using the relative motion and rotation of the camera determined for each pair of images from the time-indexed inertial sensor readings;
5. e. Reconstructing the HS cube by populating a three-dimensional data array with measured intensity values from the spatio-spectral images using the determined shift transformations.

A secondary claim is directed to the design of a static, spatio-spectral lens. The subclaims specify advantageous configurations of the method and the device.

An inertial sensor is used here to refer to a device that keeps track of changes in position and/or orientation over time and outputs them quantitatively as an electronic signal. For example, acceleration sensors and gyro sensors are such inertial sensors.

The method according to the invention can be carried out particularly advantageously with a commercially available smartphone. A common smartphone has an electronic camera, a display for captured images and video sequences (frame rate ≥ 25 Hz), an integrated inertial sensor for capturing translational and rotational movements, an electronic data memory and a usually powerful computing unit as a microprocessor, all of which are structurally compact are combined in one device unit. The arithmetic unit is programmable, and the method steps according to the invention c. to e. can be executed automatically by a program - an app.

A first essential aspect of the invention is the use of a static, spatio-spectral camera attachment, which is described in more detail below. This makes it possible in the first place to use the compact smartphone technology without any conversion effort and reversibly at any time in order to also carry out hyperspectral image acquisition.

A second essential aspect of the invention is that in method step b. no trajectory or rotation has to be predetermined. Although the method can also be used with exactly known movements of the camera - eg by a robot arm - smartphones are usually held in the hand and guided by the user. Knowing the camera movement performed during image acquisition from the inertial sensor measurement values simplifies the assignment of image information of the same object acquired sequentially in time quite considerably. This is already known from computer vision and augmented reality, see eg DE 10 2005 045 973 A1 , and is applied here for the first time to hyperspectral imaging.

The free choice of position and free mobility of the camera require that an HS cube of an object to be determined is redefined each time an image sequence is recorded. Before the start of image acquisition, the HS cube is initialized in the computing unit as an empty three-dimensional data array with initially undefined coordinate intervals.

In a preferred embodiment of the method according to the invention, the coordinate intervals are then determined by the first image of the continuously acquired sequence of spatio-spectral images of the object or the scene, i.e. the first acquired spatio-spectral image is viewed as the diagonal surface of the HS cube and is thereby defined the corner points of the HS cube, namely the coordinate intervals for image coordinates and wavelengths in which the data array is to be assigned intensity values. The image coordinates of the first image are thereby reinterpreted as quasi-object coordinates, to which all images of the sequence recorded later are to be related with their respective image coordinates. By moving and rotating the camera during image acquisition, new wavelength information can be continuously acquired for each object point that is located within the initially set framework of quasi-object coordinates. Object points outside of this frame are then not part of the HS cube determination and do not have to be processed. In this respect it is important that during image acquisition the object or scene - now specifically selected by the first image of the sequence and described by the two-dimensional interval of quasi-object coordinates - remains at least partially in the camera's field of view. If images are captured that do not meet this condition, they can be removed from the image sequence without harm. The measurement data of the inertial sensor can also be used to sort out those images that can no longer show the object captured in the first image of the sequence due to the camera position and/or viewing direction.

It can be advantageous if the object or scene is illuminated with light having a previously known spectral distribution. Smartphones in particular usually have a built-in LED that can serve as the main light source in a darkened room, for example. As already mentioned, it can also happen that objects or scenes show a too weak contrast or otherwise unfavorable response to the lighting. A remedy can then preferably be created by arranging at least one previously known reference marker on the object or in the scene, which emits a previously known spectral response when illuminated with the predetermined spectral distribution.

Algorithms for matching image coordinates from images of an object recorded from different perspectives using image structures are known per se. The intensity values of each recorded image can be assigned from the image coordinates to object coordinates—here the quasi-object coordinates—by means of determinable displacement transformations. In the case of the spatio-spectral camera, the recorded wavelength is also defined by the image coordinates, so that the intensity values can be transferred to the empty, initialized three-dimensional data array.

It can be advantageous if the data array, which is finite in size in practical programming and includes a natural number of data entries, is initialized with values of -1 in order to identify the intensity value 0 as a valid assignment. This means that entries with a value of 1 can be recognized as data gaps in the data array to be occupied. This can be of particular interest if the reconstruction of the hyperspectral cube preferably takes place in step with the acquisition of the spatio-spectral images and the inertial sensor measurement values. The real-time reconstruction requires the independent availability of information about the movement of the camera during image acquisition and is therefore within the scope of the possibilities of the method according to the invention. At the same time, the real-time reconstruction allows the generation of feedback to the user on how completely he has determined the searched HS cube. Preferably, instructions for moving and/or tilting the camera can even be calculated from existing data gaps in the three-dimensional data array by the computing unit, the execution of which captures spatio-spectral images from which the data gaps are filled. These instructions can be shown to the user as arrows on the smartphone display, for example. He then has the opportunity to immediately complete his measurement of the HS cube by moving the camera accordingly.

• 1 eine Skizze des optischen Aufbaus mit einer Darstellung der Wellenlängenselektion des von einem Objektpunkt ausgehenden Lichts im zugehörigen Bildpunkt;
• 2 eine Skizze des optischen Aufbaus mit zwei verschiedenen Wellenlängen ausgehend von zwei Objektpunkten, die gleichzeitig abgebildet werden.

In order to carry out the method according to the invention with a smartphone, a static, spatio-spectral camera attachment is required, which is to be arranged in front of the camera lens installed in the smartphone. Its optical structure is after explained in more detail below with reference to figures. It shows:

• 1 a sketch of the optical structure with a representation of the wavelength selection of the light emanating from an object point in the associated pixel;
• 2 a sketch of the optical setup with two different wavelengths starting from two object points that are imaged simultaneously.

1. a. eine Fokussierlinse und eine Kollimierlinse hintereinander angeordnet, so dass die Linsen eine gemeinsame Brennebene besitzen,
2. b. eine in der gemeinsamen Brennebene der Linsen angeordnete Schlitzblende mit einem vorbestimmten Schlitzverlauf,
3. c. ein erstes dispersives optisches Element mit Dispersionsrichtung senkrecht zum Schlitzverlauf angeordnet vor der Fokussierlinse,
4. d. ein zweites dispersives optisches Element mit Dispersionsrichtung senkrecht zum Schlitzverlauf angeordnet hinter der Kollimierlinse.

In principle, it is proposed as a camera attachment for capturing spatio-spectral images to be designed as a rigid arrangement with at least the following optical components along the optical path:

1. a. a focusing lens and a collimating lens arranged one behind the other so that the lenses have a common focal plane,
2. b. a slit diaphragm arranged in the common focal plane of the lenses with a predetermined slit course,
3. c. a first dispersive optical element with the direction of dispersion perpendicular to the course of the slit arranged in front of the focusing lens,
4. i.e. a second dispersive optical element with the direction of dispersion perpendicular to the course of the slit is arranged behind the collimating lens.

It is understood here that the light emanating from each object point should arrive at the first dispersive optical element collimated with an angle of incidence dependent on the position of the object point. This usually requires an upstream lens.

In the 1 and 2 the dispersive optical elements are shown as transmission gratings T ₁ and T ₂ . It is further assumed for the sake of simplicity that the lenses have no chromatic aberration.

In 1 the light emanating from the object point x is collimated by the upstream lens L ₁ and guided onto the grating T ₁ . The line direction of the grating T ₁ is parallel to the slit profile of the slit diaphragm S, ie the light passing through the grating T ₁ is spectrally broken down perpendicularly to the slit profile. The focusing lens L ₂ focuses all wavelengths into the focal plane with the slit S, but only the red light also passes through the slit S to the collimating lens L ₃ . The collimated red light is finally deflected again by diffraction when passing through the grating T ₂ , which, like the grating T ₁ , has the dispersion direction perpendicular to the course of the slit, and directed to the image point x' by the camera lens L ₄ , which is set to infinity. Only the wavelength λ _r arrives there.

The light from all object points with image coordinate x can pass through the slit diaphragm S in the same way, ie the entire object point line at x is imaged onto the image point line at x'. The image structure parallel to the course of the slit is retained, only the wavelength λ _r for the entire image line is selected.

In 2 the case is sketched in which blue light λ _b starting from the object point x ₁ is focused in the image point x ₁ ', while at the same time red light λ _r from the object point x ₂ enters the image point x ₂ '. A rainbow color image is generated in relation to image lines, in which the image line index encodes both the local position of the object point line and the wavelength emitted from there, in other words a spatio-spectral image.

It is known that red wavelengths are deflected (by diffraction) more than blue ones by transmission gratings. Instead of gratings, one can also use prisms as dispersive elements in which this is exactly the opposite (due to refraction). The grating constants and prism shapes determine how far the wavelengths are dispersed and thus also the width of the spectral range that can be captured in the image or the spectral resolution. It is therefore considered advantageous if the rigid arrangement under a. until d. mentioned components is designed as a modular system, so that at least the dispersive optical elements are designed as interchangeable modules. In this way, a wide variety of configurations of the camera attachment are possible with just a few simple steps, e.g. the use of transmission gratings with any number of different grating constants or combinations of gratings and prisms.

Finally, it should be noted that the person skilled in the camera attachment according to the 1 and 2 the close relationship, for example, to the double-grating monochromator US4455087A or to other spectroscopes with a central slit diaphragm. Alone, as a rigid structure without moving parts and the possibility of wavelength scanning that this gives, the camera attachment is not really usable for capturing an HS cube. In principle, it could be used on a flying drone by Sami et al., but since the camera attachment according to the invention cannot be miniaturized at will, a more compact solution is chosen there that can no longer be represented as a simple camera attachment. And remarkably, even Kim et al. not occurred to when converting a smartphone for multispectral image acquisition, exploit the mobility of the smartphone during image capture.

Claims (8)

A method of determining a hyperspectral (HS) cube of an object or scene, comprising the steps of: a. Providing a mobile electronic camera with a static, spatio-spectral camera attachment and at least one inertial sensor rigidly connected to the camera; b. Moving the mobile camera along a trajectory and/or rotating the camera about at least one camera axis such that the object or the scene remains at least partially in the camera's field of view, wherein c. continuously time-indexed spatio-spectral images and inertial sensor readings are captured by the camera while moving and/or twisting; i.e. determining displacement transformations from pairs of time-indexed images using the relative motion and rotation of the camera determined for each pair of images from the time-indexed inertial sensor readings; e. Reconstructing the HS cube by populating a three-dimensional data array with measured intensity values from the spatio-spectral images using the determined shift transformations. procedure after claim 1 , characterized in that the coordinate intervals of the three-dimensional data array are predetermined by the first image of the continuously recorded sequence of spatio-spectral images of the object or the scene. Method according to one of the preceding claims, characterized in that the object or the scene is illuminated with light having a previously known spectral distribution. procedure after claim 3 , characterized in that at least one known reference marker is arranged on the object or in the scene, which emits a known spectral response when illuminated with the predetermined spectral distribution. Method according to one of the preceding claims, characterized in that the reconstruction of the hyperspectral cube takes place in step with the acquisition of the spatio-spectral images and the inertial sensor measurement values. procedure after claim 5 , characterized in that instructions for moving and/or tilting the camera are calculated from existing data gaps in the three-dimensional data array, the execution of which captures spatio-spectral images from which the data gaps are filled. Camera attachment for capturing spatio-spectral images designed as having a rigid arrangement a. a focusing lens and a collimating lens arranged one behind the other so that the lenses have a common focal plane, b. a slit diaphragm arranged in the common focal plane of the lenses with a predetermined slit course, c. a first dispersive optical element with the direction of dispersion perpendicular to the course of the slit arranged in front of the focusing lens, i.e. a second dispersive optical element with the direction of dispersion perpendicular to the course of the slit is arranged behind the collimating lens. Camera attachment after claim 7 , characterized in that at least the dispersive optical elements are designed as interchangeable modules.

Images