METHOD AND SYSTEM FOR OBJECT RECOGNITION VIA A COMPUTER VISION APPLICATION

Abstract

A method and system for object recognition via a computer vision application including an object to be recognized, the object having an object specific luminescence spectral pattern, a light source including at least two illuminants for illuminating a scene including the object to be recognized by switching between the two illuminants, a sensor configured to capture radiance data of the scene including the object when the scene is illuminated by the light source, and a data storage unit storing fluorescence spectral patterns together with appropriately assigned respective objects. The method and system further include a data processing unit configured to extract the object specific fluorescence spectral pattern from the radiance data of the scene and to match the extracted object specific fluorescence spectral pattern with the fluorescence spectral patterns stored in the data storage unit, and to identify a best matching fluorescence spectral pattern and its assigned object.

Claims

1. A system for object recognition via a computer vision application, the system comprising at least the following components: an object to be recognized, the object having object specific reflectance and luminescence spectral patterns, a light source which is composed of at least two illuminants and is configured to illuminate a scene including the object to be recognized by switching between the at least two illuminants, wherein at least one of the at least two illuminants is based on at least one solid-state lighting system, a sensor which is configured to measure radiance data of the scene including the object when the scene is illuminated by the light source, a data storage unit which comprises luminescence spectral patterns together with appropriately assigned respective objects, a data processing unit which is configured to extract the object specific luminescence spectral pattern of the object to be recognized out of the radiance data of the scene and to match the extracted object specific luminescence spectral pattern with the luminescence spectral patterns stored in the data storage unit, and to identify a best matching luminescence spectral pattern and, thus, its assigned object.

2. The system according to claim 1, further comprising a display unit which is configured to display at least the identified object which is assigned to the identified best matching luminescence spectral pattern.

3. The system according to claim 1, wherein the object to be recognized is imparted with a predefined luminescence material and the resulting object's luminescence spectral pattern is known and used as a tag.

4. The system according to claim 1, wherein the at least one solid-state system is chosen from the group of solid-state systems comprising semiconductor light-emitting diodes (LEDs), organic light-emitting diodes (OLEDs), or polymer light-emitting diodes (PLEDs).

5. The system according to claim 1, wherein the data processing unit is configured to identify the best matching luminescence spectral pattern by using any number of matching algorithms between the extracted object specific luminescence spectral pattern and the stored luminescence spectral patterns, the matching algorithms being chosen from the group comprising at least: lowest root mean squared error, lowest mean absolute error, highest coefficient of determination, matching of maximum wavelength value.

6. The system according to claim 1, wherein the processing unit is configured to estimate, using the measured radiance data under the at least two illuminants, the luminescence spectral pattern and the reflective spectral pattern of the object to be recognized.

7. The system according to claim 1, wherein the sensor is a hyperspectral camera or a multispectral camera.

8. The system according to claim 1, wherein the sensor has one or more narrow bandpasses that correspond to Fraunhofer lines.

9. The system according claim 1, wherein the light source is a switchable light source with two illuminants each comprised of one or more LEDs and with a short switchover time between the two illuminants.

10. The system according to claim 1, wherein the sensor is synchronized to the switching of the light source to only measure at one time the radiance data from the scene under one of the at least two illuminants.

11. The system according to claim 1 wherein the sensor is synchronized to the light source and that the sensor tracks the illuminants'status during the sensor integration time.

12. The system according to claim 1 wherein the fluorescence spectral patterns of the objects which are stored in the data storage unit are coupled to information about the respective objects, particularly about type of material, price, manuals, and to other dynamic information of interest held at the data storage unit that is configured to track and update the information in 3D maps dynamically.

13. A method for object recognition via a computer vision application, the method comprising at least the following steps: providing an object with object specific reflectance and luminescence spectral patterns, the object is to be recognized illuminating a scene including the object with a light source which is composed of at least two illuminants, by switching between the at least two illuminants, wherein at least one of the two illuminants is based on at least one solid-state system, measuring, by means of a sensor, radiance data of the scene including the object when the scene is illuminated by the light source, providing a data storage unit with luminescence spectral patterns together with appropriately assigned respective objects, estimating, by a data processing unit, the object specific luminescence spectral pattern of the object to be recognized out of the radiance data of the scene, and matching, by the data processing unit, the estimated object specific luminescence spectral pattern with luminescence spectral patterns stored in the data storage unit, and identifying, by the data processing unit, a best matching luminescence spectral pattern and, thus, its assigned object.

14. The method according to claim 13, wherein the step of providing an object to be recognized comprises imparting the object with a luminescence material, thus providing the object with object specific reflectance and luminescence spectral patterns.

15. The method according to claim 13, further comprising the following step: displaying via a display device at least the identified object which is assigned to the identified best matching luminescence spectral pattern.

16. The method according to claim 13, wherein the matching step comprises to identify the best matching specific luminescence spectral pattern by using any number of matching algorithms between the estimated object specific luminescence spectral pattern and the stored luminescence spectral pattern, the matching algorithms being chosen from the group comprising at least: lowest root mean squared error, lowest mean absolute error, highest coefficient of determination, matching of maximum wavelength value.

17. The method according to claim 13, wherein the estimating step comprises to estimate, using the measured radiance data under the at least two illuminants, the luminescence spectral pattern and the reflective spectral pattern of the object in a multistep optimization process.

18. The method according to claim 13, wherein the light source is chosen as a switchable light source with two illuminants each comprised of one or more LEDs and with a short switchover time between the two illuminants.

19. A computer program product having instructions that are executable by a computer, the computer program product comprising instructions: to provide an object with object specific reflectance and luminescence spectral patterns, the object is to be recognized to illuminate a scene including the object with a light source which is composed of at least two illuminants, by switching between the at least two illuminants, wherein at least one of the two illuminants is based on at least one solid-state system, to measure, by means of a sensor, radiance data of the scene including the object when the scene is illuminated by the light source, to provide, by a data storage unit, luminescence spectral patterns together with appropriately assigned respective objects, to estimate, by a processing unit, the object specific luminescence spectral pattern of the object to be recognized out of the radiance data of the scene, and to match, by the processing unit, the estimated object specific luminescence spectral pattern with luminescence spectral patterns stored in the data storage unit, and to identify a best matching luminescence spectral pattern and, thus, its assigned object.

20. The computer program product according to claim 19 further comprising instructions: to display at least the identified object which is assigned to the identified best matching luminescence spectral pattern.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0081] FIGS. 1A and 1B show schematically embodiments of the proposed system.

[0082] FIG. 2 shows measured radiances of three broadband light sources used in example 1.

[0083] FIG. 3 shows an example comparison of measured and calculated emission spectral patterns for one material from example 1.

[0084] FIG. 4 shows measured (top) and calculated (bottom) emission spectra (spectral patterns) for all materials for example 1.

[0085] FIGS. 5A-5C show in different tables different comparing/matching algorithms for example 1.

[0086] FIG. 6 shows a diagram of example illuminant spectrums and measured radiances under a LED light source used in an embodiment of the proposed system.

[0087] FIG. 7 shows a diagram of an example comparison of measured and calculated emission spectra (spectral patterns).

[0088] FIG. 8 shows calculated emission spectra (spectral patterns) (left) and measured emission spectra (spectral patterns) (right).

[0089] FIGS. 9A-9C show in different tables different comparing/matching algorithms which can be used for example 2.

DETAILED DESCRIPTION

[0090] FIG. 1A and FIG. 1B show schematically embodiments of the proposed system 100. The system 100 includes at least one object 130 to be recognized. Further the system 100 includes a sensor 120 which can be realized by an imager, such as a camera, particularly a multispectral or a hyperspectral camera. The system 100 further includes a light source 110. The light source 110 is composed of different individual illuminants, the number of which and nature thereof depend on the method used. For example 1 as indicated in FIG. 1A, three illuminants are provided and the three illuminants are commonly available incandescent 111, compact fluorescent 112, and white light LED 113 bulbs. The light source may also be composed of two illuminants as shown in FIG. 1B. For example 2, only two illuminants are provided, the two illuminants are custom LED illuminants 114 and 115. Illuminant 114 consists of three LEDs operating at 5 V. One LED is a 400 nm LED from VCC (VAOL-5GUV0T4), with an inline resistor of 3300 ohms. The second LED is a 500 nm LED from Lumex (SSL-LX5093UEGC), with an inline resistor of 3300 ohms. The third LED is a 610 nm LED from Lumex (SSL-LX5094SOC), with an inline resistor of 680 ohms. Illuminant 115 consists of three LEDs operating at 5V. One LED is a 470 nm LED from Cree, Inc. (C503B-BCS-CV0Z0461), with an inline resistor of 5000 ohms. The second LED is a 574 nm LED from Kingbright (WP7113CGCK), with an inline resistor of 100 ohms. The third LED is a 643 nm LED from VCC (VAOL-5GAE4), with an inline resistor of 47 ohms.

[0091] The light source may be configured to illuminate a scene including the object 130 to be recognized by rapidly switching between the different illuminants (111, 112 and 113 in FIG. 1A, or 114 and 115 in FIG. 1B). The system 100 further comprises a data processing unit, i.e. CPU 140 which is configured to estimate an object specific reflectance and/or fluorescence spectral pattern with reflectance and/or fluorescence spectral patterns stored in a data storage unit 150 which is wirelessly or over a wire connected with the CPU 140 and to identify a best matching reflectance and/or fluorescence spectral pattern and, thus, an object which is assigned to the best matching reflectance and/or fluorescence spectral pattern. The system 100 further includes a display unit 160 which is configured to display at least the identified object which is assigned to the identified best matching fluorescence spectral pattern. The system 100 can comprise more than one sensor, more than one light source and/or more than one object wherein all components of the system form a network. Using the system 100 it is possible to perform an embodiment of the proposed method for object recognition. The object 130 to be recognized is imparted with a fluorescent material, thus providing the object with an object specific reflectance and fluorescence spectral pattern. To create materials with unique fluorescent spectral patterns/signatures, BASF Lumogen F series dyes dispersed into a one-component automotive clear coat were used. Four different dyes, including Yellow 170, Orange 240, Pink 285, or Red 305, labelled A, B, C, and D, and a combination of Yellow 170 and Orange 240, labelled E, was used to create materials for examples 1 and 2. The tinted clear coats were drawn down onto white steel panels at 200 micron thickness and cured. Other methods of applying fluorescence to an object may be used.

[0092] The coated object 130 is illuminated with the light source 110 which is composed of multiple illuminants. The illuminants may be rapidly switched at a rate that is not visible to the human eyes and the illuminant changes managed by the proposed system through the network, working in sync with the integration times of the sensor 120. Generally, it is possible that multiple light sources connected to the network can be synced to have the same temporal and spectral change frequencies amplifying the effect. When the scene including the object 130 is illuminated by the light source 110 radiance data of the scene including the object 130 are captured/measured by the sensor 120. The data processing unit 140 estimates the object specific reflectance and/or fluorescence spectral pattern out of the radiance data of the scene by first separating fluorescence and reflectance spectra of the object.

[0093] Multiple methods of separating fluorescence from reflectance are known. The method used in example 1 is described in Yinqiang Zheng, Imari Sato, and Yoichi Sato, Spectra Estimation of Fluorescent and Reflective Scenes by Using Ordinary Illuminates, ECCV 2014, Part V, LNCS 8693, pp. 188-202, 2014. The method described therein images a fluorescent material under three different broadband illuminants with a hyperspectral camera. This paper in incorporated by reference in full.

[0094] According to the present invention, using the measured radiance data under three different illuminants 111, 112, and 113 as shown in FIG. 1A, the reflectance and fluorescence spectral patterns are calculated in a multistep optimization process. The calculated fluorescence spectral pattern for the object 130 is compared by the processing unit 140 to the known and measured (using a fluorometer) spectral pattern from a library of materials stored in the database 150. The database 150 includes multiple fluorescence spectral patterns linked with specific objects, respectively. To form such a database it is possible to design different fluorescent formulations and applying those fluorescent formulations on respective different objects so that each object is uniquely linked with an object specific fluorescence spectral pattern. The fluorescent formulations can be designed by using specific mixtures of fluorescent chemicals with different emission profiles, in specific ratios to achieve unique spectral signatures, respectively. The fluorescent material applied to the object 130 can then be identified by any number of matching algorithms between the calculated object specific fluorescence spectral pattern and the known material spectral patterns stored in the database 150, for example, by lowest root mean squared error, lowest mean absolute error, highest coefficient of determination, or matching of maximum emission wavelength value. Identification of the fluorescent material then allows for the identification of object 130 using the database information.

[0095] Finally, the data processing unit 140 matches the estimated fluorescence spectral pattern with object-specific fluorescence spectral patterns stored in the data storage unit 150 and identifies the best matching fluorescence spectral pattern. Finally, the data processing unit 140 can read out from the data storage unit 150 by means of the identified best matching fluorescence spectral pattern the object which is linked to this best matching fluorescence spectral pattern and can display the object together with the fluorescence spectral pattern on the display unit 160.

[0096] The imager 120 can be a hyperspectral camera or a multispectral sensor. Instead of the two dozen or more individual sensor bands in a hyperspectral sensor, a multispectral sensor has approximately 4 to 20 sensor bands. Multispectral sensors can operate in snapshot mode, capturing an entire scene during a single exposure. In contrast, hyperspectral sensors typically operate in line scanning mode, meaning they cannot image the entire scene at one time. Additionally, multispectral sensors are much more economical than hyperspectral cameras. Multispectral sensors do not have the same spectral resolution as hyperspectral cameras, but they are sufficient to predict the material identification using the proposed method with appropriate matching algorithms. The sensor may also operate in a monochrome manner, with a mechanism to change the spectral region measured through time. The sensor may operate with narrow-band filters. This may be useful in outdoor conditions or other conditions with a solar lighting component when the narrow-band filters correspond to Fraunhofer lines, which are wavelengths missing from the solar spectrum due to elemental absorption within the sun. In this manner, the solar radiation, which may be overpowering compared to the artificial light source, can largely be excluded, allowing for the separation of reflectance and fluorescence and therefore object identification.

[0097] The fluorescent object 130 was imaged under the different illuminants, 111, 112, and 113 for example 1 as indicated in FIG. 1A, or the LED illuminants 114 and 115 for example 2 as indicated in FIG. 1B. The sensor 120 used is a Resonon Pika L hyperspectral imager for examples 1 and 2, consisting of 300 wavelength bands between approximately 384 nm and 1024 nm and positioned roughly 0.5 metres from the object 130. For example 2, the resulting radiances where rebinned to 10 nm intervals between 420 nm and 740 nm.

[0098] The method used to separate fluorescence from reflectance used in example 2 is in the paper of Fu et al. Separating Reflective and Fluorescent Compenents Using High Frequency Illumination in the Spectral Domain, ICCV 2013. As applied in their paper, the method requires customizable light source (Nikon ELS-VIS) capable of outputting a sinusoidal-like spectrum. The customizable light source is low powered and expensive, preventing widespread use or use in typically sized scenes. Surprisingly, it has been found here that the light source can be replaced with inexpensive and high-powered LEDs despite current LED technology being unable to create as narrow of emission bands as the Nikon ELS-VIS. The hyperspectral images were recorded in the same manner as Example 1 and rebinned to 10 nm intervals. Wavelengths at which both LED illuminants 114, 115 have similar radiances are omitted due to the nature of the calculation. The calculated/estimated emission results were compared with the fluorescence emission measured for each material using a fluorescence spectrophotometer. To facilitate easy comparison, the measured emission spectrum was also rebinned to the same 10 nm intervals and the same wavelengths omitted.

[0099] For achieving the calculated/estimated emission results, a simple algorithm is applied to the measured radiance data at each wavelength under each illuminant of the two LED illuminants 114, 115 and thus allows for separation of the reflectance and fluorescence emission spectra to be captured.

[0100] Since reflection and fluorescence have different physical behaviours, they need to be described by different models. The radiance of a reflected surface depends on the incident light and its reflectance. Using the nomenclature of the above mentioned paper Fu et al., the observed radiance of an ordinary reflected surface at a wavelength is computed as

p.sub.r()=l().Math.r()(1)

where l() is the spectrum of the incident light at wavelength and r() is the spectral reflectance of the surface at wavelength .

[0101] The observed radiance of a pure fluorescent surface depends on the incident light, the material's absorption spectrum, and its emission spectrum. Fluorescence typically absorbs light at some wavelengths and emits them at longer wavelengths. The surface's absorption spectrum will determine how much of the light is absorbed. Some of the absorbed energy is then released in the form of an emission spectrum at longer wavelengths than the incident light. The remainder of the absorbed energy is released as heat. The observed spectrum of pure fluorescent surface at wavelength is described in terms of its absorption and emission spectra as

p.sub.f(l()a()d)e()(2)

where a()and e() represent the absorption and emission spectrum. With k=I()a()d ), p.sub.f() can be written as p.sub.f()=ke() which means that the shape or the distribution of the emitted spectrum is constant but the scale k of the emitted spectrum changes under different illuminations. Namely, the radiance of the fluorescent emission changes under different illuminations, but its colour stays the same regardless of illumination colour. Finally, the reflective and fluorescent surface shows a radiance according to:

p()=l().Math.r()+ke()(3)

[0102] When using, as proposed according to the proposed method, high frequency sinusoidal illuminance in the spectral domain, the radiance of the object under these two sinusoidal illuminants can be described as:

p.sub.1()=l.sub.1()r()+ke()

p.sub.2()=l.sub.2()r()+ke()(4)

[0103] Therefrom, the reflectance r() and the fluorescent emission ke() can be recovered as

[00001]$\begin{array}{cc}r\left(\right)=\frac{p_1\left(\right)-p_2\left(\right)}{l_1\left(\right)-l_2\left(\right)}& \left(5\right)\\ \mathrm{ke}\left(\right)=p_1\left(\right)-\frac{p_1\left(\right)-p_2\left(\right)}{l_1\left(\right)-l_2\left(\right)}.Math.l_1\left(\right)& \left(6\right)\end{array}$

[0104] By means of the above described equations it is possible to calculate from the radiance data p() and the intensity I() from the illuminants the reflectance r() and the fluorescent emission e() of the object which has been illuminated by the light source. Thereby, the fluorescent emission corresponds to the object specific fluorescence spectral pattern of the object. The calculated object specific fluorescence spectral pattern is then compared with the fluorescence spectral patterns which are stored in the database and linked with respective specific objects.

[0105] FIG. 2 shows example illuminant spectra 230, 240, and 250. The diagram 200 shows a horizontal axis 210 along which the wavelength is plotted and a vertical axis 220 shows the intensity of the illumination. The curve 230 shows the illumination of a first illuminant, namely a CFL (Compact Fluorescent Lamp) with at least three pronounced maxima, namely at 435.15 nm, at 546.47 nm and the highest maximum at 611.45 nm. The curve 240 shows the illuminant spectrum of a second illuminant, namely an incandescent illuminant with a light increase in intensity with increasing wavelength. The curve 250 shows the illuminant spectrum of a third illuminant, namely a LED with two pronounced maxima, namely at 453.54 nm and at 603.02 nm.

[0106] FIG. 3 shows a comparison of calculated emission results (calculated fluorescence spectral patterns) for example 1 with fluorescent emissions measured for material A using a fluorometer. The diagram 300 is spanned over a horizontal axis 310 showing the wavelength and a vertical axis 320 showing the normalized emission intensity. As can be seen from the curves 330 and 340 wherein the curve 330 shows the calculated emission with a maximum at 565.26 nm and the curve 340 shows the measured emission with a maximum at 568 nm, a good accordance is visible.

[0107] FIG. 4 shows on the top portion measured emission spectra (measured fluorescence spectral patterns) and on the bottom portion respective calculated emission spectra (calculated fluorescence spectral patterns) for different materials A, B, C, D, E, for example 1. In each diagram for each different material A, B, C, D, E a different curve is plotted as indicated.

[0108] FIGS. 5A-5C show the results of quantitative comparisons between the calculated and measured fluorescence emission spectra for example 1. The mean absolute error (FIG. 5A), spectral angle (FIG. 5B) and Euclidean distance (FIG. 5C) were calculated for every calculated spectrum in relation to the spectra for each measured material. The mean absolute error is a common method of comparing the error of a calculated value to the ground truth value, lower mean absolute error values indicate a better match between the calculated and ground truth values. Spectral angle mapping (FIG. 5B) is a concept used in spectral imaging to classify objects to a known database of spectra. For spectral angle mapping, a lower value is indicative of a closer match between the unknown object and the measured object. Euclidean distance (FIG. 5C) is another concept used in spectral imaging in the same manner as spectral angle. Again, lower values indicate a better match for Euclidean distance. For the materials A, C, D and E the mean absolute error, spectral angle, and Euclidean distance calculation results correctly identify the unknown material, with the exception of material B, as can be seen from the tables shown in FIG. 5A, FIG. 5B and FIG. 5C, respectively.

[0109] FIG. 6 shows example illuminant spectrums and measured radiances under each illuminant for example 2. The diagram 600 shows a horizontal axis 610 along which the wavelength is plotted and a vertical axis 620 that shows the intensity of the illumination. The curve 640 shows the illumination of a first illuminant and the curve 641 shows the respective radiance data of the object illuminated by the first illuminant. The curve 630 shows the illuminant spectrum of a second illuminant and the curve 631 shows the respective radiance data of the object when illuminated by the second illuminant. The effect of fluorescence emission is obvious in the range from 530-650 nm.

[0110] FIG. 7 shows a comparison of calculated emission results (calculated fluorescence spectral patterns) for example 2 with fluorescence emissions measured for material A using a fluorometer. The diagram 700 is spanned over a horizontal axis 710 showing the wavelength and a vertical axis 720 showing the normalized emission intensity. As can be seen from the curves 730 and 740 wherein the curve 730 shows the calculated emission and the curve 740 shows the measured emission, a good accordance is visible.

[0111] FIG. 8 shows on the left side calculated emission spectra (calculated fluorescence spectral patterns) for example 2 and on the right side respective measured emission spectra for different materials A, B, C, D, E. In each diagram for each different material A, B, C, D, E a different curve is plotted as indicated.

[0112] FIGS. 9A-9C show the results of quantitative comparisons between the calculated and measured fluorescence emission spectrum for example 2. The mean absolute error (FIG. 9A), spectral angle (FIG. 9B) and Euclidean distance (FIG. 9C) were calculated for every calculated spectrum in relation to the spectra for each measured material. For each of the materials A, B, C, D and E the mean absolute error, spectral angle, and Euclidean distance calculation results correctly identify the unknown material as can be seen from the tables shown in FIG. 9A, FIG. 9B and FIG. 9C.