Classifications

• • 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
  • G06T7/55—Depth or shape recovery from multiple images
  • G06T7/593—Depth or shape recovery from multiple images from stereo images
• • 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
  • H04N13/239—Image signal generators using stereoscopic image cameras using two 2D image sensors having a relative position equal to or related to the interocular distance
• • 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
  • H04N13/257—Colour aspects
• • 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
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
  • H04N2013/0074—Stereoscopic image analysis
  • H04N2013/0081—Depth or disparity estimation from stereoscopic image signals

Definitions

• the present invention relates to the field of 3D reconstruction of a scene, especially when it is captured using asynchronous sensors.
• Asynchronous event-based vision sensors deliver compressed digital data as events.
• Event-based vision sensors have the advantage of removing redundancy, reducing latency, and increasing the range of time dynamics and gray levels compared to conventional cameras.
• the output of such a vision sensor may consist, for each pixel address, in a sequence of asynchronous events representative of changes in reflectance of the scene as they occur.
• Each pixel of the sensor is independent and detects changes in light intensity greater than a threshold since the emission of the last event (for example a contrast of 15% on the logarithm of the intensity). When the intensity change exceeds the set threshold, an ON or OFF event is generated by the pixel as the intensity increases or decreases (DVS sensors).
• a threshold since the emission of the last event (for example a contrast of 15% on the logarithm of the intensity).
• an ON or OFF event is generated by the pixel as the intensity increases or decreases (DVS sensors).
• Some asynchronous sensors associate detected events with absolute light intensity measurements (ATIS sensors).
• the senor Since the sensor is not sampled on a clock like a conventional camera, it can account for the sequencing of events with a very high temporal precision (for example of the order of 1 ⁇ ). If we use such sensor to reconstruct a sequence of images, one can reach a frame rate of several kilohertz, against a few tens of hertz for conventional cameras.
• the present invention aims to improve the situation.
• the present invention proposes a method specially adapted to asynchronous sensors for reconstructing scenes observed in 3D.
• the present invention thus aims at a 3D reconstruction method of a scene, the method comprising:
• first asynchronous information from a first sensor having a first pixel matrix arranged with respect to the scene, the first asynchronous information comprising, for each pixel of the first matrix, first successive events coming from said pixel; receiving a second asynchronous information from a second sensor having a second matrix of pixels arranged with regard to the scene, the second asynchronous information comprising, for each pixel of the second matrix, second successive events from said pixel, the second sensor being distinct from the first sensor;
• the cost function comprises at least one of:
• luminance component being a function of at least:
• motion component being a function of at least:
• temporal values relating to the occurrence of spatially localized events at a predetermined distance from a pixel of the second sensor are temporal values relating to the occurrence of spatially localized events at a predetermined distance from a pixel of the second sensor.
• the cost function can further include:
• temporal component being a function of a difference between:
• the cost function may furthermore comprise:
• geometrical component said geometrical component being a function:
• the luminance signal of the pixel of the first sensor and the pixel of the second sensor comprising a maximum encoding a time of occurrence of a luminance variation
• the convolution core may be a Gaussian of predetermined variance.
• said luminance component can be further function:
• said motion component can be further function:
• said motion component can be function, for a given time:
• said motion component can be function:
• the present invention also provides a device for 3D reconstruction of a scene, the method comprising:
• a processor adapted for pairing a first event among the first successive events with a second event among the second successive events according to a minimization of a cost function; in which the cost function comprises at least one of:
• luminance component being a function of at least:
• motion component being a function of at least:
• temporal values relating to the occurrence of spatially localized events at a predetermined distance from a pixel of the second sensor are temporal values relating to the occurrence of spatially localized events at a predetermined distance from a pixel of the second sensor.
• a computer program implementing all or part of the method described above, installed on a pre-existing equipment, is in itself advantageous.
• the present invention also relates to a computer program comprising instructions for implementing the method described above, when this program is executed by a processor.
• This program can use any programming language (for example, a language object or other), and be in the form of a source code interpretable, partially compiled code or fully compiled code.
• Figure 6 described in detail below can form the flow chart of the general algorithm of such a computer program.
• FIG. 1 is a block diagram of an asynchronous light sensor of the ATI S type
• FIG. 2 is a diagram showing events generated by an asynchronous sensor placed opposite a scene comprising a rotating star;
• FIG. 3 is an example of the calculation of a luminance component for two points of two distinct sensors
• FIGS. 4a and 4b are examples of representation of an activity signal of a given pixel
• FIG. 4c is a representation of motion maps generated using separate asynchronous sensors
• FIGS. 5a and 5b are representations of examples of calculation of geometric components in one embodiment of the invention.
• FIG. 6 illustrates a flow chart showing an embodiment according to the invention
• FIG. 7 illustrates a device for implementing an embodiment according to the invention.
• Figure 1 illustrates the principle of ATIS.
• a pixel 101 of the matrix constituting the sensor has two elements photosensitive 102a, 102b, such as photodiodes, respectively associated with electronic detection circuits 103a, 103b.
• the sensor 102a and its circuit 103a produce a pulse P 0 when the light intensity received by the photodiode 102a varies by a predefined amount.
• the pulse P 0 marking this intensity change triggers the electronic circuit 103b associated with the other photodiode 102b.
• This circuit 103b then generates a first pulse P and a second pulse P 2 as soon as a given amount of light (number of photons) is received by the photodiode 102b.
• the time difference between the pulses Pi and P 2 is inversely proportional to the light intensity received by the pixel 101 just after the appearance of the pulse P 0 .
• the asynchronous information from the ATIS comprises two combined pulse trains for each pixel (104): the first pulse train P 0 indicates the times when the light intensity has changed beyond the detection threshold, while the The second train consists of the pulses Pi and P 2 , the time difference of which indicates the corresponding light intensities, or gray levels.
• An event e (p, t) coming from a pixel 101 of position p in the matrix of the ATIS then comprises two types of information: a temporal information given by the position of the pulse P 0 , giving the instant t of the event, and gray level information given by the time difference between pulses Pi and P 2 .
• each point p identifies an event e (p, t) generated asynchronously at a given instant.
• t at a pixel p of the sensor, of position ar the movement of a star rotating at constant angular velocity as
• the majority of these points are distributed in the vicinity of a generally helical surface.
• the figure shows a number of events at a distance from the helical surface that are measured without corresponding to the actual motion of the star. These events are acquisition noise.
• the events e (p, t) can then be defined by all of the following information: with C the spatial domain of the sensor, pol the polarity representing the direction of the luminance change (eg 1 for an increase or -1 for a decrease) and I (p, t) the luminous intensity signal from point p to l moment t.
• the intensity signal of the pixel situated at the coordinate p then makes it possible to code the luminance information temporally. This information can be directly from the electronic circuit of the sensor with a minimum of transformation.
• FIG. 3 is an example of the calculation of a luminance component for two points p and q of two distinct u and v sensors.
• the surfaces composing the observed scene are Lambert surfaces (ie surfaces whose luminance is the same as whatever the angle of observation).
• the luminance component can be expressed as: I u (Pi, t) I v ( ⁇ li, t) dt
• the core g a (t) is advantageously a Gaussian of variance ⁇ . It can also be a gate function of width ⁇ .
• FIG. 4a shows three possible activity signals t ⁇ S, for three pixels pi, p 2 and p 3 of the sensor (and for a given value of polarity pol).
• S (pi, t), S (p 2 , t) or S (p 3 , t) is zero.
• S (pi, t) takes a predetermined threshold value (here h, this value h may be unitary).
• the value of the activity signal S (p-i, t) then decreases gradually after this event to tend towards 0.
• the value of the activity signal S can be set to the sum (possibly weighted) of the current value of S just before the event 422 (ie 0 ) and h.
• the decay of the curve S will start from the value h + h 0 as shown in FIG . 4b .
• the value of the curve S is set to the value h regardless of the value of h 0 (ie the events prior to the last event (ie the subsequent event) are ignored).
• the last event time defined as follows:
• T (p, pol, i) max (t / ) ⁇ j ⁇ i or
• T (p, pol, t) max (t / )
• p ⁇ T p, pol, t defines a map of the times of the last events of the same polarity occurring temporally just before a reference time (i.e. t).
• p ⁇ S (p, pol, t) is a function of this set of times T (p, pol, t).
• p ⁇ S (p, pol, t) tT p, pol, t)
• a predetermined time constant ⁇ S may be any function decreasing with time t over an interval comprising as lower bound T ⁇ p, pol, t)).
• a pixel map S representative of the "coolness" of events of these pixels is advantageous because it allows a continuous and simple representation of discontinuous concepts (i.e. events). This created map makes it possible to transform the representation of the events in a simple domain of apprehension.
• This S function is representative of a "freshness" of events occurring for this pixel.
• the cards 401 and 402 of FIG. 4c are representations of this function S for a given time t and for two asynchronous sensors capturing the movement of the same hand from two different points of view.
• the darkest points represent points whose latest events are the most recent with respect to time t (i.e. having the largest value of S).
• the clearest points represent points whose last events are the farthest from time t (ie, having the smallest value of S, the background of the image is grayed out to make it easier to highlight the clear values, although the background corresponds to null values of the function S).
• the scattered dark spots correspond to capturing noise from the sensors.
• each pixel p of the card has the value S (p, t 0 ).
• the set of points near p define a set v u (p) (405) and the set of points close to q define a set v v (g) (406) ) (N is the cardinal of these sets).
• the motion component for a given instant t, can be expressed as follows:
• FIGS. 5a and 5b are representations of examples of calculation of geometric components in one embodiment of the invention.
• R u is the projection center of the sensor 501 and R v is the projection center of the sensor 502.
• This epipolar line l uv is defined as the intersection of the plane (X (t), R u , R v ) with the sensor 502.
• a point p of the first sensor 501 defines an epipolar line 1 v (p) on the second sensor 502 and a point q of the second sensor 502 defines an epipolar line 1 u (q) on the first sensor 501.
• a point p of the first sensor and a point q of the second sensor define an epipolar intersection i w (p, q) on the third sensor;
• a point p of the first sensor and a point r of the third sensor define an epipolar intersection i v (p, r on the second sensor;
• a point q of the second sensor and a point r of the third sensor define an epipolar intersection i u (q, r) on the first sensor.
• the camera has more than three sensors (eg Q sensors), it is possible to generalize the previous formula by considering that the epipolar intersection of a sensor is the point closest to all the sensors. Epipolar lines defined on this sensor by the current points of the other sensors (for example, minimizing the sum of the distances or minimizing the square of the distances of the said points to the epipolar lines).
• Figure 6 illustrates a flow chart showing an embodiment according to the invention.
• step 603 On receiving two sets of asynchronous events 601 and 602 coming from two different asynchronous sensors and facing the same scene, it is possible to select two events of these sensors (step 603, defined by the pixel p t and the time t u for the first sensor and the pixel qj and the time t 2 j for the second sensor).
• a temporal component (step 605)
• a luminance component (step 607).
• the calculated distances (or the position of the point X (t) in the space) are then returned (612).
• FIG. 7 illustrates a device for implementing an embodiment according to the invention.
• the device comprises a computer 700, comprising a memory 705 for storing instructions for implementing the method, the received measurement data, and temporary data for performing the various steps of the method as described above. .
• the computer further comprises a circuit 704.
• This circuit can be, for example:
• processor capable of interpreting instructions in the form of a computer program
• a programmable electronic chip such as an FPGA (for "Field Programmable Gate Array”).
• This computer has an input interface 703 for receiving the events of the sensors, and an output interface 706 for the delivery of the distances 707.
• the computer can comprise, to allow easy interaction with a user, a screen 701 and a keyboard 702.
• the keyboard is optional, especially in the context of a computer in the form of a touch pad, for example.
• FIG. 6 is a typical example of a program whose instructions can be implemented with the device described. As such, FIG. 6 may correspond to the flowchart of the general algorithm of a computer program within the meaning of the invention.
• the flowchart of FIG. 6 may also include an iteration over the events e ⁇ t ⁇ ) to associate several events of the first sensor with events of the second sensor.

Landscapes

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

Priority Applications (5)

Applications Claiming Priority (2)

Publications (1)

Family

ID=53879566

Family Applications (1)

Country Status (7)

Cited By (2)

Families Citing this family (7)

Family Cites Families (12)

• 2015
  • 2015-03-16 FR FR1552154A patent/FR3033973A1/fr active Pending
• 2016
  • 2016-03-15 US US15/556,596 patent/US11335019B2/en active Active
  • 2016-03-15 CN CN201680016173.XA patent/CN107750372B/zh active Active
  • 2016-03-15 KR KR1020177029337A patent/KR102432644B1/ko active Active
  • 2016-03-15 EP EP16713966.6A patent/EP3272119B1/fr active Active
  • 2016-03-15 JP JP2017549009A patent/JP6839091B2/ja not_active Expired - Fee Related
  • 2016-03-15 WO PCT/FR2016/050575 patent/WO2016146938A1/fr not_active Ceased

Non-Patent Citations (4)

Also Published As

Similar Documents

Legal Events