Classifications

• • G—PHYSICS
  • G01—MEASURING; TESTING
  • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
  • G01B11/00—Measuring arrangements characterised by the use of optical techniques
  • G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
  • G01B11/25—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures by projecting a pattern, e.g. one or more lines, moiré fringes on the object
  • G01B11/2518—Projection by scanning of the object
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
  • H04N13/20—Image signal generators
  • H04N13/204—Image signal generators using stereoscopic image cameras
  • H04N13/25—Image signal generators using stereoscopic image cameras using two or more image sensors with different characteristics other than in their location or field of view, e.g. having different resolutions or colour pickup characteristics; using image signals from one sensor to control the characteristics of another sensor
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T5/00—Image enhancement or restoration
  • G06T5/20—Image enhancement or restoration using local operators
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T5/00—Image enhancement or restoration
  • G06T5/70—Denoising; Smoothing
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/50—Depth or shape recovery
  • G06T7/521—Depth or shape recovery from laser ranging, e.g. using interferometry; from the projection of structured light
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
  • H04N13/20—Image signal generators
  • H04N13/296—Synchronisation thereof; Control thereof
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20036—Morphological image processing
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20112—Image segmentation details
  • G06T2207/20164—Salient point detection; Corner detection

Definitions

• the present invention is related to a three dimensional imaging system.
• One of the state-of-the-art implementations of depth imaging system is based on structures or lines projected on the scene, which are imaged by an imaging system.
• a fixed line is projected on a conveyor belt and the objects to be scanned are passing through the laser curtain.
• a high-speed imaging system (order of 1 kps frame rate) is monitoring the distortions of the projected line and deducting the depth profile of the object where the line is projected.
• the speed of the scan along the object is defined by the speed of the conveyor which is typically slow with respect to the imaging system framerate.
• This invention formulates an optimized sensor to maximize performance of the laser line scanning system, by being able to image a line using an exposure of only 10’s of nanoseconds while minimizing the optical power requirements even in ambient light conditions.
• a first aspect of the present invention is related to a method for determining the depth profile of a field of view comprising the steps of: o projecting by means of a projector (5) at least one light pattern onto the field of view, the projection occurring in a time window of less than 10psec, preferably less than 1 psec ; o imaging the projected pattern by means of a camera sensor (4) and optics, synchronised with the projection of the pattern during at least one observation window within said time window, said camera sensor comprising a matrix of pixels (1 ), each comprising a photodetector, said pixels (9) being in a false status when no light is detected by the corresponding photodetector, and in a true status when light is detected by the corresponding photodetector thereby obtaining a first binary matrix of pixels representing the field of view; o separating the projected pattern from an ambient light noise (10) on the binary matrix of pixels by considering only pixels in the true status having at least one neighbour pixel also in the true status on the matrix of pixel (1
• Preferred embodiments of the method of the invention comprise one or a suitable combination of more than one of the following features:
• the projected pattern comprises at least one continuous line, the step of separating the projected pattern from the ambient noise by only considering true pixels forming at least one continuous line;
• the projected continuous line is a straight line, sequentially scanned on the entire field of view or on a partial part of it forming a region of interest;
• the straight line is oriented in a predetermined direction, the step of separating line from the ambient noise using the predetermined direction for analysing a probability of a neighbour pixel being part of the projected straight line;
• the probability of a pixel being part of the projected pattern is determined by a trained neural network
• the displacement of the projected pattern between two successive projections is corresponding to less than one pixel width, the depth resolution being improved by interpolation of the determined depth between successive line scans.
• the projected line is produced by moving at least one laser beam on the field of view;
• each photodetector comprises a single photon detector.
• a second aspect of the invention is related to a depth imaging system implementing the method of the invention:
• an imaging device comprising a matrix of pixels each pixel comprising a photodetector able to detect single photons impinging thereon and optics able to make an image of the field of view on the matrix of pixel, said single photon detector having a binary logic status of true when a photon is detected and a logic status of false when no photon is detected in a timeframe;
• a projector able to project a pattern in a time window of less than 10psec, preferably less than 1 psec;
• the projector is arranged to project the lines in a predetermined direction for improving line detection by the logic.
• the matrix of pixel comprises a first layer comprising the single photon detector and a second layer comprising a neighbourhood testing circuit, the neighbourhood testing circuit being arranged so that the output of the second layer circuit being true only if the corresponding single photon detector and at least one adjacent single photon detector are simultaneously in true status, thereby determining which pixels have at least one contiguous pixel in the true logic status.
• Figure 1 represents an ultra-fast line scanner according to the invention.
• Figure 2 a to d represents lines can pixelisation in line scan spatial dithering conditions.
• Figure 3 represents the results of interpolation of the position of the line scan of figure 2 along the line A-A.
• Figure 4 represents results of random spatial dithering to estimate subpixel resolution.
• Figure 5 represents results of estimate subpixel resolution by known dot movement.
• Figure 6 represents line scan interleaved motion.
• Figure 7 and 8 represents projections of different patterns instead of a simple line.
• Figure 9 represents a two layer imaging sensor.
• Image sensor (pixel matrix)
• imaging plane such as a diffraction grating
• the invention described below proposes a system that is able to image a 3 dimensional scene at high speeds.
• the system projects a light pattern such as a line on the scene at high frequency, the pattern being displaced between each shot to be able to reconstruct the entire scene.
• a 1-axis oscillating MEMS projecting a laser line, sweeping this laser line across the field of view (FoV) at 1 kHz to 10kHz, meaning that in a period of 10Ous to 1 ms the complete scene or FoV is sensed.
• each line needs to be imaged with the imaging system each meaning 100ns to 1 us per line (Tline)
• the invention is related to depth imaging system comprising a sensor, said sensor comprising an array of pixels, each pixel being capable of detecting single photons, for example by means of single-photon avalanche diode (SPAD).
• a sensor comprising an array of pixels, each pixel being capable of detecting single photons, for example by means of single-photon avalanche diode (SPAD).
• SPAD single-photon avalanche diode
• the sensor is preferably synchronised with a pulsing laser line (or pattern), so that the sensor knows when to expect photons from each projected laser line.
• the pixels are activated during a short observation window, preferably ranging from one nanosecond to 100s of nanoseconds. During this observation window, the single photon detector pixels trigger or change state when a photon is detected. Photons detected will not only originate from the laser line energy, but also from ambient light present in the scene.
• the region of interest is read out column by column or row by row, and the pixel map is built to identify the location and shape of the line. Since we know the line is continuous and piece- wise continuous, this can be used as a constraint to identify which pixels are part of the line. For example, an isolated pixel event is likely not part of a piece-wise continuous structure.
• another constraint can be that when considering a vertical line as projected structure, in the detected ROI map, 2 separate events in the same row, must be either adjacent - in which case the line is detected on the boundary of or partially covering the adjacent pixels, in which case we can apply some super-resolution - or in the case the detected events are not adjacent, only 1 of the pixels will have a line-connected event and the other pixel will have a random event. By looking at the events in the next or previous row, 1 can get additional information as to which event was the true event, since we expect true events to be connected in piece-wise linear fashion in the vertical dimension - since we are projecting a vertical line.
• any orientation of the line could be applied if one does not have this limitation, for example this is applicable in massively parallel connected processing in the case of stacked systems, where the pixel plane is connected in a parallel fashion to the processing layer, rather than by busses such as typically is the case in image sensors.
• a second aspect of the invention is related to a stack devices, whereby a first layer includes the photon detectors such as the SPAD array as well as some optional electronics to control the SPAD array and pixels, possibly quenching resistors or circuits, and whereby a second layer, parallel ly connected to the first layer by means of 1 or more electrical connections per detector or detector group.
• the second layer processes the information coming the first layer.
• the coincidence imaging as well as the line detection is performed by logic in the second layer.
• a massive parallel connection connects the SPAD array with a neural network which computes the location of the line segments in the raw data image, whereby the neural network is trained to distinguish between noise and line information.
• FIG. 36 Figure shows a sequence of lines in the case of horizontal resolution of 10 lines each oscillation period of the line scanner.
• the laser line is moving from left to right, which we can call the sweep-right lines, while from T5 to T9 the laser line is moving from right to left, the sweep-left lines, in this figure in an interleaved fashion, creating a resolution of 10 in the horizontal direction each period of the line scanner.
• the pixel array in its entirety or a region of interest of the array is connected in parallel to a process plane.
• the pixel array or sensor array which is sensitive to the optical input can be positioned on a first plane, with each pixel connected by means of an interconnect method to the processing layer physically located on a second plane of the sensor.
• the delta depth is linear with the delta disparity. This means that the quantization of the disparity limits the accuracy of the depth estimation.
• the limitation of the disparity quantization because of pixel pitch is mitigated, by means of a spatial dithering.
• a spatial dithering By constructing the depth estimation from a collection of neighboring, related measurements at slightly different locations in the scene, the depth estimation can be improved. This idea is similar to the idea of quantization with a 1 -bit AD convertor and dithering. If the dithering noise is white and surpass the quantization, by means of averaging of the end results, a better estimate can be created. The same applies to this spatial dithering concept, whereby the pixel is quantizer and the moving dot is the spatial dither. If the spatial dither surpass the pixel quantization in terms of amplitude, and the distribution of the dither is known (it is, because the system controls the light) a more accurate subresolution measurement can be obtained.
• the invention builds further on the possibility of extremely fast sensing - almost continuous - , whether it is a line, a dot or a dot pattern.
• Figure 4 shows a location dithering method.
• the pixel array will sample this function on discrete moments in time using pixels 303 with finite size creating a quantization.
• This quantization creates a histogram 314.
• an average location 315 can be computed from the histogram 314, which is more accurate than if the dot with a static location would be observed by the quantizing pixels. In such a way, the accuracy of the average dot location can be obtained with greater resolution.
• a third aspect of the invention is related to a method for determining the depth profile of a field of view, similar to the method according to the first aspect.
• step c (separating) is optional, since the method of determining the depth profile may also be performed with other filtering steps as understood by the skilled person.
• a fourth aspect of the invention is related to a method for determining the depth profile of a field of view, similar to the method according to the first aspect.
• step b comprises imaging by means of two cameras, each having a matrix of pixels
• step d comprises triangulation based on the positions of the two cameras and the two matrices of pixels.
• triangulating data of two cameras or of one camera and one light source is conceptually similar.
• a fifth aspect of the invention is related to a depth imaging system, preferably implementing the method of the third and fourth aspects.
• the projector projects a simple structure such as an ellipsoid or circular dot onto the scene and scans the scene in a raster, Lissajous or another pattern.
• each camera is adapted to track the position of the dot in each of the image planes and output a stream of (x, y, t) data where x, y are coordinates in the image plane (pixel dimension) and t is the timestamp of the detection.
• x, y are coordinates in the image plane (pixel dimension) and t is the timestamp of the detection.
• t is the timestamp of the detection.
• Fig. 10 shows the laserbased illumination continuously sweeps the world in fast strokes.
• a Lissajous pattern is generated based on a fast 2D MEMS mirror.
• Two or more sensors snapshot the position of the laser dot at a very fast rate (up to 100 MHz). Each sensor sends out the laser dot position, after which a simple triangulation algorithm allows to compute precise 3D shape, position, contour, and motion at said very fast rate.
• the methods and systems of this invention are advantageous for a few reasons as summarized below.
• the first advantage is related to power.
• a single dot (or a set of dots) is projected at a time, bundling all the energy to sample a certain location in the scene in the shortest time possible, obtaining the best immunity to ambient light for the lowest photon budget.
• the required photon budget to be able to sense the dot depends on the sensing technology used. For example, it is possible to detect a dot with just 12 photons, distributed on 2 pixels (6 photons/pixel). Furthermore, the shorter the time needed to produce a return of 6 photons per pixel, the better the immunity to ambient light. Therefore, this system preferably uses short high power pulses.
• the system in nearfield regions, can increase pulse density and reduce pulse intensity to achieve a more precise mesh. In more distant regions, the system can reduce pulse density and increase pulse intensity to achieve a useable signal.
• the product of pulse density and intensity can be maintained under class-1 eye safe constraints if desired, or could be further reduced at times when power consumption concerns outweigh scan density needs. The scan pattern does not need to be altered to produce these varying- density scans.
• a second advantage of this invention is related to bandwidth. Scanning depth sensing systems offer a unique advantage in the way the depth data is represented and transmitted.
• the depth data can be represented as a series of x, y, z points, as expected from any point cloud representation. However, chronologically, the points in the point cloud are created along a defined laser beam trajectory. As such, the points along the trajectory are typically ‘connected’ in a certain way, influenced by the scene itself. The fact that the points or samples are connected along a trajectory can be leveraged in the data representation. Instead of outputting each sample individually, the trajectory can be described with a limited set of coefficients and 2 points describing the start and stop of the polynomial descriptor.
• a third advantage of this invention is related to immunity.
• Laser beam scanning based systems using the sensors of this invention offer best-in- class immunity to concurrent active systems.
• the depth information is obtained by triangulating single point pairs in the respective image spaces tracking the dot.
• the introduction of additional systems in the environment will create additional dots scanning the scene.
• the sensors used to track the dot are capable of tracking multiple dots, the additional optical information created by the additional scanning dot, will just contribute to the depth data being obtained as another point pair that can be triangulated.
• Laser beam scanning systems have also immunity to flood systems (active stereo, mono NIR with flood, iToF).
• the system is capable of suppressing the influence from ambient light by imposing requirements on the photon statistics and behaviour of the projected dot.
• Flood illumination resembles in terms of statistics and behaviour to ambient light: it does create high peak flux (energy/pixel/time) and requires typically several hundreds of microseconds to milliseconds to get some integrated signal in the pixels, while not having any spatio-temporal structure.
• the system uses a combination of photon statistics, beam shape, and beam spatio-temporal movement to distinguish signal from ambient /flood sources and optical signal origination from the laser beam scanning system itself.
• Laser beam scanning systems have also immunity to dot pattern projectors.
• Two elements distinguish a Structured Light system using dot patterns from dynamic Laser beam scanning structured light. Firstly, the dot pattern is static and will not have any moving behaviour or trajectories. Secondly, the available optical power is spread between 10k or so dot in parallel, meaning that the allocated peak power per dot is significantly less than the allocated peak power in the Laser beam scanning system. Thirdly, the structured light projectors are designed to work together with global shutter imagers, which have typical exposure times in the order of 100s of microseconds. The energy required to see the dots is thus spread in the exposure window. [67] A fourth advantage of this invention is related to latency.
• the system uses a fully serialized sensing method, where the data is produced serially as the beam scans the field of view.
• the time needed between the optical sampling of the scene (projection of a dot on a certain surface) and the production of the depth data corresponding to that sample is very short, since the method relies purely on the following.
• the sensing of the dot location in the different image spaces using this system of this invention can be in the order of 10 nanoseconds.
• the transfer of the coordinates to the sensing host N clock-cycles, typically 100’s of nanoseconds.
• filtering method using historical data and triangulation of the point in space which can be pipelined math operations in 100’s of nanoseconds.
• Fig. 15 shows (a) the ground truth, (b) raw detections, and (c) filtered detections.
• a first filtering step is happening inside the pixel plane, meaning this filter operates in parallel on each of the pixels.
• this filter exhibits either a spatial filtering method, a temporal filtering method, or a combination thereof.
• These filters can be implemented using transistor level, RTL, more complex compute architectures, neural networks or other.
• the system relies on two basic principles for filtering in the pixel plane.
• the first principle is spatial. By imposing a certain shape on the projected dot we can constrain the pixel to only flag a valid detection when the impose kernel was detected within a specified time. For example, a 2x1 kernel can be imposed, meaning that the projected dot must span at least 2x1 pixels.
• the constraint imposed on the pixels will be that a cluster of 2x1 pixels must fire within tp in order to consider the detected photons to come from the active projection. This principle is referred to as coincidence.
• a second principle is spatio-temporal. We can impose that the trajectory of the dot is continuous, provided the pulse repetition frequency is higher than the dot displacement velocity. For example, in case of 1 ns pulses repeated every 40ns, a temporal window of 41 ns will be able to catch 2 pulses. By imposing that the pixel (or pixel cluster) must see at least 2 pulses in the 41 ns window we can filter out a significant portion of the detected ambient photons, since the ambient light photon statistics have a low chance to generate 2 consecutive detections in s 42ns window. This principle is referred to as persistence. By applying a combination of both coincidence and persistence, a powerful filtering strategy is created which is quite efficient in terms of hardware implementation.
• the fastest way to output the array data is to compress it, for example by means of projection onto the axis of the imager.
• the imager is first divided, for example, into 16x16 tiles of 64x64 pixels whereby each of the tiles can output its projections to the digital logic typically on the periphery of the device.
• the projection operation creates 2 vectors (times NxN tiles), 1 vector each for the row and the column projections.
• the creation of the vectors based on the projection data is the step which takes the data from the asynchronous domain inside the pixel array to the synchronous domain at the periphery. While the inpixel filtering can be implemented synchronously, it is not required. As such, the vectorization of the projections is discretizing the data in time, for instance every 10ns a new vector is created based on the projection events coming in during those 10ns.
• the displacement of the projected pattern between two successive projections is corresponding to less than one pixel width, the depth resolution being improved by interpolation of the determined depth between successive line scans.
• this is not to be understood as limited to line scans, as said interpolation can work equally well when using dot scans.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Optics & Photonics (AREA)
• Length Measuring Devices By Optical Means (AREA)

Priority Applications (6)

Applications Claiming Priority (2)

Related Child Applications (2)

Publications (1)

Family

ID=78649091

Family Applications (1)

Country Status (6)

Cited By (2)

Families Citing this family (2)

Citations (4)

Family Cites Families (5)

• 2021
  • 2021-10-04 EP EP21200661.3A patent/EP4160140A1/en not_active Withdrawn
• 2022
  • 2022-10-03 CN CN202280062865.3A patent/CN117999457A/zh active Pending
  • 2022-10-03 WO PCT/EP2022/076182 patent/WO2023057208A1/en not_active Ceased
  • 2022-10-03 KR KR1020247014252A patent/KR20240090268A/ko active Pending
  • 2022-10-03 US US18/694,914 patent/US12445586B2/en active Active
  • 2022-10-03 EP EP22789902.8A patent/EP4237792B1/en active Active
  • 2022-10-03 JP JP2024519427A patent/JP2024535455A/ja active Pending
• 2025
  • 2025-09-03 US US19/318,260 patent/US20260067438A1/en active Pending

Patent Citations (4)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events