WO2022025741A1 - Estimation de profondeur basée sur un réseau - Google Patents

Estimation de profondeur basée sur un réseau Download PDF

Info

Publication number
WO2022025741A1
WO2022025741A1 PCT/KR2021/095070 KR2021095070W WO2022025741A1 WO 2022025741 A1 WO2022025741 A1 WO 2022025741A1 KR 2021095070 W KR2021095070 W KR 2021095070W WO 2022025741 A1 WO2022025741 A1 WO 2022025741A1
Authority
WO
WIPO (PCT)
Prior art keywords
map
disparity
reference image
image frame
values
Prior art date
Application number
PCT/KR2021/095070
Other languages
English (en)
Inventor
Chenchi Luo
Yingmao Li
Kaimo Lin
Youngjun Yoo
Original Assignee
Samsung Electronics Co., Ltd.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US17/027,106 external-priority patent/US11816855B2/en
Application filed by Samsung Electronics Co., Ltd. filed Critical Samsung Electronics Co., Ltd.
Publication of WO2022025741A1 publication Critical patent/WO2022025741A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06T7/596—Depth or shape recovery from multiple images from stereo images from three or more stereo images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N13/00—Stereoscopic video systems; Multi-view video systems; Details thereof
    • H04N13/10—Processing, recording or transmission of stereoscopic or multi-view image signals
    • H04N13/106—Processing image signals
    • H04N13/128—Adjusting depth or disparity
    • 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
    • H04N13/243—Image signal generators using stereoscopic image cameras using three or more 2D image sensors
    • 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
    • H04N13/271—Image signal generators wherein the generated image signals comprise depth maps or disparity maps
    • 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

  • This disclosure relates generally to imaging systems. More specifically, this disclosure relates to array-based depth estimation.
  • Many mobile electronic devices such as smartphones and tablet computers, include cameras that can be used to capture still and video images.
  • multiple cameras can be used to simultaneously capture multiple images of scenes, such as when left and right cameras of an electronic device are used to simultaneously capture two images of a scene.
  • the ability to simultaneously capture multiple images of a scene allows an electronic device to perform disparity processing in order to identify depths of different image pixels within the scene.
  • Disparity refers to the difference in pixel locations of the same point in a scene as captured in different images of the scene.
  • Depth has a known relationship to disparity. A point within a scene that is farther away (has a larger depth) will typically have a smaller disparity, meaning pixels capturing that point in different images will be closer to each other in the images.
  • a point within a scene that is closer (has a smaller depth) will typically have a larger disparity, meaning pixels capturing that point in different images will be farther apart from each other in the images.
  • a method comprises obtaining a plurality of image frames of a scene captured using a plurality of imaging sensors, the plurality of image frames comprising a reference image frame and a plurality of non-reference image frames; generating a first disparity map based on the reference image frame and a first non-reference image frame among the plurality of non-reference image frames; generating a second disparity map based on the reference image frame and a second non-reference image frame among the plurality of non-reference image frames; generating, based on the first disparity map, a first confidence map comprising a first plurality of confidence levels associated with a first plurality of disparity values of the first disparity map; generating, based on the second disparity map, a second confidence map comprising a second plurality of confidence levels associated with a second plurality of disparity values of the second disparity map; and generating a depth map of the scene based on the first disparity map, the second disparity map, the first confidence map and the second confidence map, wherein the
  • FIGURE 1 illustrates an example network configuration including an electronic device in accordance with this disclosure
  • FIGURE 2 illustrates an example imaging array for use with array-based depth estimation in accordance with this disclosure
  • FIGURE 3A illustrates example disparities in image frames captured using the imaging array of FIGURE 2 in accordance with this disclosure
  • FIGURE 3B illustrates example disparities in image frames captured using the imaging array of FIGURE 2 in accordance with this disclosure
  • FIGURE 3C illustrates example disparities in image frames captured using the imaging array of FIGURE 2 in accordance with this disclosure
  • FIGURE 4 illustrates an example technique for array-based depth estimation in accordance with this disclosure
  • FIGURE 5 illustrates an example machine learning-based architecture for array-based depth estimation in accordance with this disclosure
  • FIGURE 6 illustrates an example technique for cross-correlation to support array-based depth estimation in accordance with this disclosure
  • FIGURE 7 illustrates an example technique for confidence map generation to support array-based depth estimation in accordance with this disclosure
  • FIGURE 8 illustrates an example technique for information fusion to support array-based depth estimation in accordance with this disclosure
  • FIGURE 9A illustrates example results that may be obtained using array-based depth estimation in accordance with this disclosure
  • FIGURE 9B illustrates example results that may be obtained using array-based depth estimation in accordance with this disclosure
  • FIGURE 9C illustrates example results that may be obtained using array-based depth estimation in accordance with this disclosure
  • FIGURE 10A illustrates another example imaging array for use with array-based depth estimation and related details in accordance with this disclosure
  • FIGURE 10B illustrates another example imaging array for use with array-based depth estimation and related details in accordance with this disclosure
  • FIGURE 11A illustrates yet other example imaging arrays for use with array-based depth estimation in accordance with this disclosure
  • FIGURE 11B illustrates yet other example imaging arrays for use with array-based depth estimation in accordance with this disclosure.
  • FIGURE 12 illustrates an example method for array-based depth estimation in accordance with this disclosure.
  • This disclosure relates to array-based depth estimation.
  • a method in a first embodiment, includes obtaining, using one or more processors, at least three input image frames of a scene captured using at least three imaging sensors.
  • the input image frames include a reference image frame and multiple non-reference image frames.
  • the method also includes generating, using the one or more processors, multiple disparity maps using the input image frames. Each disparity map is associated with the reference image frame and a different one of the non-reference image frames.
  • the method further includes generating, using the one or more processors, multiple confidence maps using the input image frames. Each confidence map identifies weights associated with one of the disparity maps.
  • the method includes generating, using the one or more processors, a depth map of the scene using the disparity maps and the confidence maps.
  • the imaging sensors are arranged to define multiple baseline directions, where each baseline direction extends between the imaging sensor used to capture the reference image frame and the imaging sensor used to capture a different one of the non-reference image frames.
  • an apparatus in a second embodiment, includes at least three imaging sensors and at least one processor.
  • the at least one processor is configured to obtain at least three input image frames of a scene using the at least three imaging sensors.
  • the input image frames include a reference image frame and multiple non-reference image frames.
  • the at least one processor is also configured to generate multiple disparity maps using the input image frames. Each disparity map is associated with the reference image frame and a different one of the non-reference image frames.
  • the at least one processor is further configured to generate multiple confidence maps using the input image frames. Each confidence map identifies weights associated with one of the disparity maps.
  • the at least one processor is configured to generate a depth map of the scene using the disparity maps and the confidence maps.
  • the imaging sensors are arranged to define multiple baseline directions. Each baseline direction extends between the imaging sensor used to capture the reference image frame and the imaging sensor used to capture a different one of the non-reference image frames.
  • a non-transitory computer readable medium contains instructions that when executed cause at least one processor to obtain at least three input image frames of a scene captured using at least three imaging sensors.
  • the input image frames include a reference image frame and multiple non-reference image frames.
  • the medium also contains instructions that when executed cause the at least one processor to generate multiple disparity maps using the input image frames. Each disparity map is associated with the reference image frame and a different one of the non-reference image frames.
  • the medium further contains instructions that when executed cause the at least one processor to generate multiple confidence maps using the input image frames. Each confidence map identifies weights associated with one of the disparity maps.
  • the term “or” is inclusive, meaning and/or.
  • various functions described below can be implemented or supported by one or more computer programs, each of which is formed from computer readable program code and embodied in a computer readable medium.
  • application and “program” refer to one or more computer programs, software components, sets of instructions, procedures, functions, objects, classes, instances, related data, or a portion thereof adapted for implementation in a suitable computer readable program code.
  • computer readable program code includes any type of computer code, including source code, object code, and executable code.
  • computer readable medium includes any type of medium capable of being accessed by a computer, such as read only memory (ROM), random access memory (RAM), a hard disk drive, a compact disc (CD), a digital video disc (DVD), or any other type of memory.
  • ROM read only memory
  • RAM random access memory
  • CD compact disc
  • DVD digital video disc
  • a “non-transitory” computer readable medium excludes wired, wireless, optical, or other communication links that transport transitory electrical or other signals.
  • a non-transitory computer readable medium includes media where data can be permanently stored and media where data can be stored and later overwritten, such as a rewritable optical disc or an erasable memory device.
  • phrases such as “have”, “may have”, “include”, or “may include” a feature indicate the existence of the feature and do not exclude the existence of other features.
  • the phrases “A or B”, “at least one of A and/or B”, or “one or more of A and/or B” may include all possible combinations of A and B.
  • “A or B”, “at least one of A and B”, and “at least one of A or B” may indicate all of (1) including at least one A, (2) including at least one B, or (3) including at least one A and at least one B.
  • first and second may modify various components regardless of importance and do not limit the components. These terms are only used to distinguish one component from another.
  • a first user device and a second user device may indicate different user devices from each other, regardless of the order or importance of the devices.
  • a first component may be denoted a second component and vice versa without departing from the scope of this disclosure.
  • the phrase “configured (or set) to” may be interchangeably used with the phrases “suitable for”, “having the capacity to”, “designed to”, “adapted to”, “made to”, or “capable of” depending on the circumstances.
  • the phrase “configured (or set) to” does not essentially mean “specifically designed in hardware to”. Rather, the phrase “configured to” may mean that a device can perform an operation together with another device or parts.
  • the phrase “processor configured (or set) to perform A, B, and C” may mean a generic-purpose processor (such as a CPU or application processor) that may perform the operations by executing one or more software programs stored in a memory device or a dedicated processor (such as an embedded processor) for performing the operations.
  • Examples of an “electronic device” may include at least one of a smartphone, a tablet personal computer (PC), a mobile phone, a video phone, an e-book reader, a desktop PC, a laptop computer, a netbook computer, a workstation, a personal digital assistant (PDA), a portable multimedia player (PMP), an MP3 player, a mobile medical device, a camera, or a wearable device (such as smart glasses, a head-mounted device (HMD), electronic clothes, an electronic bracelet, an electronic necklace, an electronic accessory, an electronic tattoo, a smart mirror, or a smart watch).
  • PDA personal digital assistant
  • PMP portable multimedia player
  • MP3 player MP3 player
  • a mobile medical device such as smart glasses, a head-mounted device (HMD), electronic clothes, an electronic bracelet, an electronic necklace, an electronic accessory, an electronic tattoo, a smart mirror, or a smart watch.
  • Other examples of an electronic device include a smart home appliance.
  • Examples of the smart home appliance may include at least one of a television, a digital video disc (DVD) player, an audio player, a refrigerator, an air conditioner, a cleaner, an oven, a microwave oven, a washer, a drier, an air cleaner, a set-top box, a home automation control panel, a security control panel, a TV box (such as SAMSUNG HOMESYNC, APPLETV, or GOOGLE TV), a smart speaker or speaker with an integrated digital assistant (such as SAMSUNG GALAXY HOME, APPLE HOMEPOD, or AMAZON ECHO), a gaming console (such as an XBOX, PLAYSTATION, or NINTENDO), an electronic dictionary, an electronic key, a camcorder, or an electronic picture frame.
  • a television such as SAMSUNG HOMESYNC, APPLETV, or GOOGLE TV
  • a smart speaker or speaker with an integrated digital assistant such as SAMSUNG GALAXY HOME, APPLE HOMEPOD, or AMAZON
  • an electronic device include at least one of various medical devices (such as diverse portable medical measuring devices (like a blood sugar measuring device, a heartbeat measuring device, or a body temperature measuring device), a magnetic resource angiography (MRA) device, a magnetic resource imaging (MRI) device, a computed tomography (CT) device, an imaging device, or an ultrasonic device), a navigation device, a global positioning system (GPS) receiver, an event data recorder (EDR), a flight data recorder (FDR), an automotive infotainment device, a sailing electronic device (such as a sailing navigation device or a gyro compass), avionics, security devices, vehicular head units, industrial or home robots, automatic teller machines (ATMs), point of sales (POS) devices, or Internet of Things (IoT) devices (such as a bulb, various sensors, electric or gas meter, sprinkler, fire alarm, thermostat, street light, toaster, fitness equipment, hot water tank, heater, or boiler).
  • MRA magnetic resource
  • the term “user” may denote a human or another device (such as an artificial intelligent electronic device) using the electronic device.
  • FIGURES 1 through 12 discussed below, and the various embodiments of this disclosure are described with reference to the accompanying drawings. However, it should be appreciated that this disclosure is not limited to these embodiments and all changes and/or equivalents or replacements thereto also belong to the scope of this disclosure.
  • disparity refers to the difference in pixel locations of the same point in a scene as captured in different images of the scene.
  • Depth has a known relationship to disparity. A point within a scene that is farther away (has a larger depth) will typically have a smaller disparity, meaning pixels capturing that point in different images will be closer to each other in the images. A point within a scene that is closer (has a smaller depth) will typically have a larger disparity, meaning pixels capturing that point in different images will be farther apart from each other in the images.
  • an electronic device can generate a depth map that identify the depths of the pixels within the scene.
  • the depth map may be used to support various image processing operations or other operations.
  • the electronic device may be unable to distinguish between different points of the feature-less pattern.
  • the electronic device may be unable to distinguish between different portions of the same repetitive pattern.
  • various other functions that rely on accurate disparity or depth estimations may not produce accurate results.
  • This disclosure provides techniques for array-based depth estimation.
  • multiple input image frames of a scene are captured using at least three imaging sensors of an electronic device.
  • the imaging sensors are arranged in a non-linear manner so that the image frames captured using the imaging sensors are displaced along multiple baseline directions (such as horizontally and vertically).
  • the input image frames have disparities in multiple directions.
  • a machine learning algorithm is applied to the image frames in order to generate multiple disparity maps and multiple confidence maps associated with the disparity maps.
  • Each disparity map is produced using a different pair of the image frames, and each disparity map is associated with a specific baseline direction that identifies an axis along which the two imaging sensors that captured the pair of the image frames are separated.
  • Each confidence map identifies the level of confidence that the machine learning algorithm has in the disparities identified in one of the disparity maps along the associated baseline direction.
  • the disparity maps and the confidence maps can be fused to produce a final depth map of the scene based on the input image frames.
  • FIGURE 1 illustrates an example network configuration 100 including an electronic device in accordance with this disclosure.
  • the embodiment of the network configuration 100 shown in FIGURE 1 is for illustration only. Other embodiments of the network configuration 100 could be used without departing from the scope of this disclosure.
  • an electronic device 101 is included in the network configuration 100.
  • the electronic device 101 can include at least one of a bus 110, a processor 120, a memory 130, an input/output (I/O) interface 150, a display 160, a communication interface 170, a sensor 180, and a flash 190.
  • the electronic device 101 may exclude at least one of these components or may add at least one other component.
  • the bus 110 includes a circuit for connecting the components 120-190 with one another and for transferring communications (such as control messages and/or data) between the components.
  • the processor 120 includes one or more of a central processing unit (CPU), a graphics processing unit (GPU), an application processor (AP), or a communication processor (CP).
  • the processor 120 is able to perform control on at least one of the other components of the electronic device 101 and/or perform an operation or data processing relating to communication.
  • the processor 120 may obtain input image frames and generate high-accuracy depth maps based on the input image frames.
  • the processor 120 may also perform one or more image processing operations or other operations based on the generated depth maps.
  • the memory 130 can include a volatile and/or non-volatile memory.
  • the memory 130 can store commands or data related to at least one other component of the electronic device 101.
  • the memory 130 can store software and/or a program 140.
  • the program 140 includes, for example, a kernel 141, middleware 143, an application programming interface (API) 145, and/or an application program (or “application”) 147.
  • At least a portion of the kernel 141, middleware 143, or API 145 may be denoted an operating system (OS).
  • OS operating system
  • the kernel 141 can control or manage system resources (such as the bus 110, processor 120, or memory 130) used to perform operations or functions implemented in other programs (such as the middleware 143, API 145, or application 147).
  • the kernel 141 provides an interface that allows the middleware 143, the API 145, or the application 147 to access the individual components of the electronic device 101 to control or manage the system resources.
  • the application 147 may include one or more applications that, among other things, obtain input image frames and generate high-accuracy depth maps based on the input image frames.
  • the application 147 may also include one or more applications that perform one or more image processing operations or other operations based on the generated depth maps. These functions can be performed by a single application or by multiple applications that each carries out one or more of these functions.
  • the middleware 143 can function as a relay to allow the API 145 or the application 147 to communicate data with the kernel 141, for instance.
  • a plurality of applications 147 can be provided.
  • the middleware 143 is able to control work requests received from the applications 147, such as by allocating the priority of using the system resources of the electronic device 101 (like the bus 110, the processor 120, or the memory 130) to at least one of the plurality of applications 147.
  • the API 145 is an interface allowing the application 147 to control functions provided from the kernel 141 or the middleware 143.
  • the API 145 includes at least one interface or function (such as a command) for filing control, window control, image processing, or text control.
  • the I/O interface 150 serves as an interface that can, for example, transfer commands or data input from a user or other external devices to other component(s) of the electronic device 101.
  • the I/O interface 150 can also output commands or data received from other component(s) of the electronic device 101 to the user or the other external device.
  • the display 160 includes, for example, a liquid crystal display (LCD), a light emitting diode (LED) display, an organic light emitting diode (OLED) display, a quantum-dot light emitting diode (QLED) display, a microelectromechanical systems (MEMS) display, or an electronic paper display.
  • the display 160 can also be a depth-aware display, such as a multi-focal display.
  • the display 160 is able to display, for example, various contents (such as text, images, videos, icons, or symbols) to the user.
  • the display 160 can include a touchscreen and may receive, for example, a touch, gesture, proximity, or hovering input using an electronic pen or a body portion of the user.
  • the communication interface 170 is able to set up communication between the electronic device 101 and an external electronic device (such as a first electronic device 102, a second electronic device 104, or a server 106).
  • the communication interface 170 can be connected with a network 162 or 164 through wireless or wired communication to communicate with the external electronic device.
  • the communication interface 170 can be a wired or wireless transceiver or any other component for transmitting and receiving signals, such as images.
  • the wireless communication is able to use at least one of, for example, long term evolution (LTE), long term evolution-advanced (LTE-A), 5th generation wireless system (5G), millimeter-wave or 60 GHz wireless communication, Wireless USB, code division multiple access (CDMA), wideband code division multiple access (WCDMA), universal mobile telecommunication system (UMTS), wireless broadband (WiBro), or global system for mobile communication (GSM), as a cellular communication protocol.
  • the wired connection can include, for example, at least one of a universal serial bus (USB), high definition multimedia interface (HDMI), recommended standard 232 (RS-232), or plain old telephone service (POTS).
  • the network 162 or 164 includes at least one communication network, such as a computer network (like a local area network (LAN) or wide area network (WAN)), Internet, or a telephone network.
  • the electronic device 101 further includes one or more sensors 180 that can meter a physical quantity or detect an activation state of the electronic device 101 and convert metered or detected information into an electrical signal.
  • one or more sensors 180 include multiple cameras or other imaging sensors, which may be used to capture image frames of scenes.
  • the sensor(s) 180 can also include one or more buttons for touch input, one or more microphones, a gesture sensor, a gyroscope or gyro sensor, an air pressure sensor, a magnetic sensor or magnetometer, an acceleration sensor or accelerometer, a grip sensor, a proximity sensor, a color sensor (such as a red green blue (RGB) sensor), a bio-physical sensor, a temperature sensor, a humidity sensor, an illumination sensor, an ultraviolet (UV) sensor, an electromyography (EMG) sensor, an electroencephalogram (EEG) sensor, an electrocardiogram (ECG) sensor, an infrared (IR) sensor, an ultrasound sensor, an iris sensor, or a fingerprint sensor.
  • a gesture sensor e.g., a gyroscope or gyro sensor
  • an air pressure sensor e.g., a gyroscope or gyro sensor
  • a magnetic sensor or magnetometer e.gyroscope or gyro sensor
  • the sensor(s) 180 can further include an inertial measurement unit, which can include one or more accelerometers, gyroscopes, and other components.
  • the sensor(s) 180 can include a control circuit for controlling at least one of the sensors included here. Any of these sensor(s) 180 can be located within the electronic device 101.
  • the cameras or other imaging sensors 180 can optionally be used in conjunction with at least one flash 190.
  • the flash 190 represents a device configured to generate illumination for use in image capture by the electronic device 101, such as one or more LEDs.
  • the first external electronic device 102 or the second external electronic device 104 can be a wearable device or an electronic device-mountable wearable device (such as an HMD).
  • the electronic device 101 can communicate with the electronic device 102 through the communication interface 170.
  • the electronic device 101 can be directly connected with the electronic device 102 to communicate with the electronic device 102 without involving with a separate network.
  • the electronic device 101 can also be an augmented reality wearable device, such as eyeglasses, that include one or more cameras.
  • the first and second external electronic devices 102 and 104 and the server 106 each can be a device of the same or a different type from the electronic device 101.
  • the server 106 includes a group of one or more servers.
  • all or some of the operations executed on the electronic device 101 can be executed on another or multiple other electronic devices (such as the electronic devices 102 and 104 or server 106).
  • the electronic device 101 when the electronic device 101 should perform some function or service automatically or at a request, the electronic device 101, instead of executing the function or service on its own or additionally, can request another device (such as electronic devices 102 and 104 or server 106) to perform at least some functions associated therewith.
  • FIGURE 1 illustrates one example of a network configuration 100 including an electronic device 101
  • the network configuration 100 could include any number of each component in any suitable arrangement.
  • computing and communication systems come in a wide variety of configurations, and FIGURE 1 does not limit the scope of this disclosure to any particular configuration.
  • FIGURE 1 illustrates one operational environment in which various features disclosed in this patent document can be used, these features could be used in any other suitable system.
  • FIGURE 2 illustrates an example imaging array 200 for use with array-based depth estimation in accordance with this disclosure.
  • the imaging array 200 of FIGURE 2 may be described as being used in the electronic device 101 of FIGURE 1.
  • the imaging array 200 of FIGURE 2 may represent one or more sensors 180 in the electronic device 101 of FIGURE 1.
  • the imaging array 200 may be used with any suitable device(s) and in any suitable system(s).
  • the imaging array 200 includes three imaging sensors 202, 204, and 206 that are arranged in a right-angle triangle pattern.
  • Each imaging sensor 202, 204, and 206 captures image data that is used to form image frames of scenes.
  • the actual image frames may be generated by the imaging sensors 202, 204, and 206 or by the processor 120 that receives the image data from the imaging sensors 202, 204, and 206.
  • the generated image frames may contain any suitable image-related data, such as red-green-blue (RGB) image data, luminance and chrominance (YUV) image data, or raw image data.
  • RGB red-green-blue
  • YUV luminance and chrominance
  • FIGURES 3A, 3B, and 3C illustrate example disparities in image frames captured using the imaging array 200 of FIGURE 2 in accordance with this disclosure.
  • FIGURE 3A represents a reference image frame 302 captured using the imaging sensor 202
  • FIGURE 3B represents a horizontal image frame 304 captured using the imaging sensor 204
  • FIGURE 3C represents a vertical image frame 306 captured using the imaging sensor 206.
  • a ghost object 310 in FIGURES 3B and 3C illustrates the position of the object 308 from the reference image frame 302 of FIGURE 3A.
  • FIGURE 3B there is a horizontal disparity 312 between the object 308 and the ghost object 310 along the baseline direction 208.
  • FIGURE 3C there is a vertical disparity 314 between the object 308 and the ghost object 310 along the baseline direction 210.
  • the electronic device 101 or other device can process three or more image frames of a scene and generate multiple disparity maps that identify disparities between the image frames along multiple baseline directions.
  • the electronic device 101 or other device can also generate confidence maps that identify different levels of confidence for the disparities identified in the disparity maps along the associated baseline directions.
  • the electronic device 101 or other device can further fuse this information into a highly-accurate depth map for the scene.
  • baseline directions 208 and 210 shown here and described as being used by the approaches discussed below are used for simplicity since they are orthogonal. However, any other suitable baseline directions, whether orthogonal or not, may be used. Also note that while often described as involving the use of three image frames captured using three imaging sensors, the approaches described below can be easily expanded for use with four or more imaging sensors.
  • FIGURE 2 illustrates one example of an imaging array 200 for use with array-based depth estimation
  • an imaging array may include three or more imaging sensors in any suitable arrangement, as long as the imaging sensors define multiple different baseline directions between various ones of the imaging sensors.
  • FIGURES 3A, 3B, and 3C illustrate one example of disparities in image frames captured using the imaging array 200 of FIGURE 2, various changes may be made to FIGURES 3A, 3B, and 3C.
  • the horizontal and vertical disparities 312 and 314 here can easily vary based on the actual depth of the object 308 relative to the imaging sensors 202, 204, and 206.
  • captured image frames will routinely include both horizontal and vertical disparities relative to a reference image frame.
  • FIGURE 4 illustrates an example technique 400 for array-based depth estimation in accordance with this disclosure.
  • the technique 400 of FIGURE 4 may be described as being used by the electronic device 101 of FIGURE 1, which may include the imaging array 200 of FIGURE 2.
  • the technique 400 may be used with any suitable device(s) having any suitable imaging array(s) and in any other suitable system(s).
  • the technique 400 receives and processes three input image frames 402, 404, and 406.
  • the image frame 402 represents a reference image frame, which in some embodiments may be captured using the imaging sensor 202.
  • the image frames 404 and 406 respectively represent a horizontal image frame and a vertical image frame, which in some embodiments may be captured using the imaging sensors 204 and 206.
  • Each image frame 402, 404, and 406 may have a resolution defined by a height H and a width W, so the image frames 402, 404, and 406 collectively have a resolution of (H, W, 3).
  • the image frame 402 is provided to a feature extractor 408, which processes the image frame 402 to identify a feature map 414 containing high-level features of the image frame 402.
  • the image frame 404 is provided to a feature extractor 410, which processes the image frame 404 to identify a feature map 416 containing high-level features of the image frame 404.
  • the image frame 406 is provided to a feature extractor 412, which processes the image frame 406 to identify a feature map 418 containing high-level features of the image frame 406.
  • Each feature extractor 408, 410, and 412 may represent a trained machine learning model or other algorithm for identifying features of image frames.
  • Each feature extractor 408, 410, and 412 may use any suitable technique to identify features of input image frames, such as when implemented using multiple layers of a trained convolutional neural network (CNN). Note that multiple feature extractors 408, 410, and 412 are shown here, and the same weights used for feature extraction may be shared between the feature extractors 408, 410, and 412.
  • CNN convolutional neural network
  • the feature maps 414 and 416 are processed using a cross-correlation function 420.
  • the cross-correlation function 420 uses a sliding search window along one baseline direction (such as the baseline direction 208) to identify correlations between the feature maps 414 and 416 of the image frames 402 and 404. These correlations are used later to identify how common points in a scene are captured at different pixel locations in the image frames 402 and 404, thereby identifying disparities associated with the image frames 402 and 404.
  • the feature maps 414 and 418 are processed using a cross-correlation function 422.
  • the cross-correlation function 422 uses a sliding search window along another baseline direction (such as the baseline direction 210) to identify correlations between the feature maps 414 and 418 of the image frames 402 and 406. These correlations are used later to identify how common points in the scene are captured at different pixel locations in the image frames 402 and 406, thereby identifying disparities associated with the image frames 402 and 406.
  • Each cross-correlation function 420 and 422 may represent a trained machine learning model or other algorithm for identifying correlations between features of image frames.
  • Each cross-correlation function 420 and 422 may use any suitable technique to identify correlations between features of input image frames, such as when implemented using one or more layers of a trained CNN.
  • Outputs of the cross-correlation function 420 include a set of correlated feature maps 424, which identify correlated features of the image frames 402 and 404 determined by the cross-correlation function 420 along the baseline direction 208. Multiple correlated feature maps 424 can be identified here, such as one correlated feature map 424 for each position of the sliding search window used by the cross-correlation function 420.
  • outputs of the cross-correlation function 422 include a set of correlated feature maps 426, which identify correlated features of the image frames 402 and 406 determined by the cross-correlation function 422 along the baseline direction 210.
  • Multiple correlated feature maps 426 can be identified here, such as one correlated feature map 426 for each position of the sliding search window used by the cross-correlation function 422.
  • the correlated feature maps 424 or 426 collectively have a resolution of , where , , and represents the size of the sliding search window used by the cross-correlation function 420 or 422.
  • the correlated feature maps 424 are processed by a disparity refinement function 428, which restores the spatial resolution of the correlated feature maps 424 following the cross-correlation function 420 to produce a disparity map 432 for the baseline direction 208.
  • the feature map 414 associated with the reference image frame 402 can be used by the disparity refinement function 428.
  • the disparity map 432 has a resolution of (H, W).
  • the correlated feature maps 426 are processed by a disparity refinement function 430, which restores the spatial resolution of the correlated feature maps 426 following the cross-correlation function 422.
  • the disparity refinement function 430 can be used to produce a disparity map 434 for the baseline direction 210.
  • the feature map 414 associated with the reference image frame 402 can be used by the disparity refinement function 430.
  • the disparity map 434 has a resolution of (H, W).
  • Each disparity refinement function 428 and 430 may represent a trained machine learning model or other algorithm for restoring spatial resolution of correlated feature maps to produce disparity maps.
  • Each disparity refinement function 428 and 430 may use any suitable technique to restore spatial resolution of correlated feature maps to produce disparity maps, such as when implemented using one or more layers of a trained CNN.
  • a fused depth map generation function 440 receives the disparity maps 432 and 434 and the confidence maps 436 and 438.
  • the fused depth map generation function 440 uses these inputs to produce a final depth map 442 for the scene captured in the input image frames 402, 404, and 406.
  • the fused depth map generation function 440 may scale the disparity values contained in the disparity map 432 or depth values based on the disparity values contained in the disparity map 432 using the confidence values contained in the confidence map 436.
  • the fused depth map generation function 440 may also scale the disparity values contained in the disparity map 434 or depth values based on the disparity values contained in the disparity map 434 using the confidence values contained in the confidence map 438.
  • the fused depth map generation function 440 may use the scaled disparity or depth values to identify final depth values contained in the depth map 442.
  • the same feature extractor may be used to serially process multiple input image frames to produce feature maps
  • the same cross-correlation function may be used to serially process different pairs of feature maps using different directions for its sliding search window to produce correlated feature maps
  • the same disparity refinement function may be used to serially process different correlated feature maps to produce disparity maps.
  • the feature extractor 408 is implemented using a collection of convolutional layers 502a-502d, which are used to process the input image frame 402.
  • Each convolutional layer 502a-502d applies a convolution function to its inputs in order to generate its outputs.
  • a convolutional layer 502a-502d generally represents a layer of convolutional neurons, which apply a convolution function that emulates the response of individual neurons to visual stimuli. Each neuron typically applies some function to its input values (often by weighting different input values differently) to generate output values.
  • a convolutional layer 502a-502d may be associated with an activation function, which can apply a specific function or operation to the output values from the neurons to produce final outputs of the convolutional layer.
  • the first convolutional layer 502a receives and processes the input image frame 402, and each of the remaining convolutional layers 502b-502d receives and processes the outputs from the prior convolutional layer 502a-502c.
  • the output of each convolutional layer 502a-502d has a lower resolution than its input.
  • the convolutional layer 502d outputs high-level features 414a
  • the convolutional layer 502c outputs high-level features 414b
  • the convolutional layer 502b outputs high-level features 414c.
  • the high-level features 414a-414c collectively represent the feature map 414 discussed above.
  • the high-level features 414a represent the features that are used by the cross-correlation functions 420 and 422, and the high-level features 414b-414c represent the features that are used by the disparity refinement functions 428 and 430. Note that while four convolutional layers 502a-502d are shown here, the feature extractor 408 may support any suitable number of convolutional layers.
  • feature extractors 410 and 412 may be implemented using the same arrangement of convolutional layers 502a-502d, which can operate using the same weights that are used in the feature extractor 408, to produce high-level features in the feature maps 416 and 418, respectively.
  • the cross-correlation layers 504 and 506 slide their respective search windows along different baseline directions (such as along the baseline directions 208 and 210) in order to identify correlations between the input image frames along the different baseline directions.
  • One example implementation of the cross-correlation layers 504 and 506 is shown in FIGURES 6 and 7, which are described below.
  • the disparity refinement function 428 is implemented using a collection of deconvolutional or upsampling layers 508a-508e and convolutional layers 510a-510b, which are used to restore spatial resolution to the correlated feature maps 424 and produce the disparity map 432.
  • each convolutional layer 510a-510b applies a convolution function to its inputs in order to generate its outputs.
  • Each deconvolutional or upsampling layer 508a-508e applies a deconvolution or upsampling function to its inputs in order to generate its outputs.
  • the first deconvolutional or upsampling layer 508a receives and processes the correlated feature maps 424 produced by the cross-correlation layer 504, and each of the deconvolutional or upsampling layers 508b-508c receives and processes the outputs from the prior deconvolutional or upsampling layer 508a-508b.
  • the outputs of the deconvolutional or upsampling layer 508c are provided to the convolutional layer 510a along with the high-level features 414b of the feature map 414 from the convolutional layer 502c, which allows the high-level features 414b for the reference input image frame 402 to be fed forward and concatenated with the outputs of the deconvolutional or upsampling layer 508c.
  • the disparity refinement function 428 may support any suitable number of deconvolutional or upsampling layers and any suitable number of convolutional layers.
  • the disparity refinement function 430 may be implemented using the same arrangement of layers 508a-508e, 510a-510b, which are used to restore spatial resolution to the correlated feature maps 426 and produce the disparity map 434.
  • the fused depth map generation function 440 is implemented using a fusion layer 512, which receives the disparity maps 432 and 434 from the disparity refinement functions 428 and 430 and the confidence maps 436 and 438 from the cross-correlation functions 420 and 422.
  • the fusion layer 512 uses this information to produce the final depth map 442 of the scene that is captured in the input image frames 402, 404, and 406.
  • FIGURE 8 One example technique for fusing disparity maps and confidence maps is shown in FIGURE 8, which is described below.
  • the same layers for a feature extractor may be used to serially process multiple input image frames
  • the same layer(s) for a cross-correlation function may be used to serially process different pairs of feature maps using different directions for its sliding search window
  • the same layers for a disparity refinement function may be used to serially process different correlated feature maps.
  • the high-level features 414a of the feature map 414 and the feature maps 416 and 418 are received as inputs.
  • the feature map 416 is provided to a shift function 602, which operates to shift a sliding window 604 within the feature map 416 along one baseline direction (such as the baseline direction 208).
  • the contents of the feature map 416 within the sliding window 604 are provided to a normalized correlation function 606, which also receives the high-level features 414a of the feature map 414.
  • the normalized correlation function 606 calculates a normalized correlation or cross-correlation between the contents of the feature map 416 within the sliding window 604 and the high-level features 414a of the feature map 414, thereby producing one of the correlated feature maps 424.
  • FIGURE 7 illustrates an example technique 700 for confidence map generation to support array-based depth estimation in accordance with this disclosure. More specifically, the technique 700 of FIGURE 7 illustrates one example implementation of another part of the cross-correlation functions 420 and 422 in the technique 400 of FIGURE 4. For ease of explanation, the technique 700 of FIGURE 7 may be described as being used by the electronic device 101 of FIGURE 1, which may include the imaging array 200 of FIGURE 2. However, the technique 700 may be used with any suitable device(s) having any suitable imaging array(s) and in any other suitable system(s).
  • the correlated feature maps 424 or 426 produced by the cross-correlation function 420 or 422 are used.
  • One or more operations 702 are applied to the correlated feature maps 424 or 426 in order to produce an initial confidence map 704, such as a lower-resolution confidence map.
  • the operations 702 may include a softmax operation applied along the channel direction of the correlated feature maps 424 or 426 followed by an argmax operation applied along the channel direction of the correlated feature maps 424 or 426.
  • the softmax operation generally remaps the values of the correlated feature maps 424 or 426 to a desired probability distribution, while the argmax operation returns the largest values from the remapped correlated feature maps 424 or 426.
  • An upsampling operation 706 is then performed to increase the resolution of the initial confidence map 704 to produce one of the confidence maps 436 or 438, which has a higher resolution than the initial confidence map 704.
  • FIGURE 6 illustrates one example of a technique 600 for cross-correlation to support array-based depth estimation
  • the technique 600 in FIGURE 6 implements both of the cross-correlation functions 420 and 422 from FIGURE 4. More specifically, the left half of FIGURE 6 implements the cross-correlation function 420, and the right half of FIGURE 6 implements the cross-correlation function 422.
  • the cross-correlation functions 420 and 422 may be implemented separately (such as in the different layers 504 and 506) since the only link between the cross-correlation functions 420 and 422 in FIGURE 6 is the common receipt of the high-level features 414a of the feature map 414.
  • FIGURE 7 illustrates one example of a technique 700 for confidence map generation to support array-based depth estimation
  • various changes may be made to FIGURE 7.
  • the number of correlated feature maps 424 or 426 can vary as needed or desired, and the contents of the correlated feature maps 424 or 426 and confidence maps 704 and 706 are for illustration only.
  • the fused depth map generation function 440 uses disparity maps 432 and 434 and confidence maps 436 and 438 to produce a final depth map 442 for a scene captured in input image frames 402, 404, and 406.
  • the fused depth map generation function 440 uses four values to determine each depth value in the final depth map 442, namely (i) a disparity value from the disparity map 432 or a depth value based on a disparity value from the disparity map 432, (ii) the confidence level of that disparity or depth value from the confidence map 436, (iii) a disparity value from the disparity map 434 or a depth value based on a disparity value from the disparity map 434, and (iv) the confidence level of that disparity or depth value from the confidence map 438.
  • each depth value in the final depth map 442 may be determined using these four values as follows:
  • D represents the computed depth value for the final depth map 442.
  • FIGURE 8 An example of these operations is shown in FIGURE 8, where the fused depth map generation function 440 is operating to convert two depth values 802 and 804 (which represent depth values along the two baseline directions 208 and 210) into two scaled depth values 806 and 808.
  • the scaled depth value 806 is larger than the original depth value 802, while the scaled depth value 808 is smaller than the original depth value 804. This indicates that the confidence level for the depth value 802 was higher than the confidence level for the depth value 804, so the original depth value 802 is being weighted more than the original depth value 804.
  • a final depth value 810 may then be computed based on the scaled depth values 806 and 808.
  • the scaled depth values 806 and 808 are also shown as being orthogonal. This allows the final depth value 810 to be easily calculated as described above. However, as shown below, this process can be easily modified to support the use of non-orthogonal baseline directions.
  • FIGURE 8 illustrates one example of a technique 800 for information fusion to support array-based depth estimation
  • various changes may be made to FIGURE 8.
  • the fused depth map generation function 440 may use any other suitable technique to scale disparity or depth values and combine the scaled disparity or depth values in order to produce final depth values for a scene.
  • the functions and other operations described above with reference to FIGURES 4, 5, 6, 7, and 8 can be implemented in an electronic device 101, 102, 104, server 106, or other device in any suitable manner.
  • the operations described above can be implemented or supported using one or more software applications or other software instructions that are executed by at least one processor 120 of a device.
  • at least some of the operations described above can be implemented or supported using dedicated hardware components.
  • the operations described above can be performed using any suitable hardware or any suitable combination of hardware and software/firmware instructions.
  • FIGURES 9A, 9B, and 9C illustrate example results that may be obtained using array-based depth estimation in accordance with this disclosure. These results may, for example, be obtained using the technique 400 described above.
  • FIGURE 9A an input image frame 902 of a scene is shown.
  • the input image frame 902 shows a person (whose face is obscured for privacy) standing in front of a tennis court.
  • FIGURE 9B a depth map 904 is shown for the scene and may be generated using a conventional approach based the input image frame 902 and another input image frame (such as left and right input image frames).
  • the depth map 904 suffers from various problems, such as where the conventional approach has difficulty discerning depths in an area 906 of the scene associated with a repetitive pattern.
  • a depth map 908 is shown for the same scene and may be generated using the technique 400 based on the input image frame 902 and at least two other input image frames. As can be seen here, the depth map 908 much more accurately identifies depths within the scene, including in the area of the scene associated with the repetitive pattern.
  • FIGURES 9A, 9B, and 9C illustrate examples of results that may be obtained using array-based depth estimation
  • various changes may be made to these figures.
  • these figures are merely meant to illustrate one example of the types of results that could be obtained using the approaches described in this disclosure.
  • images of scenes can vary widely, and the results obtained using the approaches described in this patent document can also vary widely depending on the circumstances.
  • FIGURES 10A and 10B illustrate another example imaging array 1000 for use with array-based depth estimation and related details in accordance with this disclosure.
  • the imaging array 1000 may be described as being used in the electronic device 101 of FIGURE 1.
  • the imaging array 1000 may represent one or more sensors 180 in the electronic device 101 of FIGURE 1.
  • the imaging array 1000 may be used with any suitable device(s) and in any suitable system(s).
  • the imaging array 1000 includes three imaging sensors 1002, 1004, and 1006. Each imaging sensor 1002, 1004, and 1006 captures image data that is used to form image frames of scenes.
  • the actual image frames may be generated by the imaging sensors 1002, 1004, and 1006 or by the processor 120 that receives the image data from the imaging sensors 1002, 1004, and 1006.
  • the generated image frames may contain any suitable image-related data, such as RGB image data, YUV image data, or raw image data.
  • the imaging sensors 1002 and 1004 are separated horizontally along a baseline direction 1008, and the imaging sensors 1002 and 1006 are separated diagonally along a baseline direction 1010. Because of the offsets of the imaging sensors 1002, 1004, and 1006 in the baseline directions 1008 and 1010, image frames captured using the imaging sensors have various levels of disparities, which depend on the depths of objects or backgrounds in the scene being imaged.
  • the cross-correlation function 420 may process features for a reference image frame captured using the imaging sensor 1002 and a first non-reference image frame captured using the imaging sensor 1004 in the same or similar manner described above (since the baseline directions 208 and 1008 are both horizontal).
  • the cross-correlation function 422 may process features for the reference image frame captured using the imaging sensor 1002 and a second non-reference image frame captured using the imaging sensor 1006 in a similar manner as described above, but the sliding window 610 used by the cross-correlation function 422 can slide in a diagonal direction corresponding the baseline direction 1010.
  • the fused depth map generation function 440 can also be modified to calculate final depth values based on non-orthogonal disparity or depth values.
  • An example of this is shown in FIGURE 10B, where two scaled depth values 1012 and 1014 may be used to calculate a depth valve 1016 for the final depth map 442. This may occur in a similar manner as described above, but the calculations can be easily adjusted to account for the non-orthogonal nature of the two scaled depth values 1012 and 1014 that correspond to the two non-orthogonal baseline directions 1008 and 1010.
  • FIGURES 11A and 11B illustrate yet other example imaging arrays 1100 and 1150 for use with array-based depth estimation in accordance with this disclosure.
  • the imaging arrays 1100 and 1150 may be described as being used in the electronic device 101 of FIGURE 1.
  • the imaging arrays 1100 and 1150 may each represent one or more sensors 180 in the electronic device 101 of FIGURE 1.
  • the imaging arrays 1100 and 1150 may be used with any suitable device(s) and in any suitable system(s).
  • the imaging array 1150 includes six imaging sensors 1152, 1154, 1156, 1158, 1160, and 1162 that are arranged in a hexagonal pattern.
  • Each imaging sensor 1152, 1154, 1156, 1158, 1160, and 1162 captures image data that is used to form image frames of scenes.
  • the actual image frames may be generated by the imaging sensors 1152, 1154, 1156, 1158, 1160, and 1162 or by the processor 120 that receives the image data from the imaging sensors 1152, 1154, 1156, 1158, 1160, and 1162.
  • the generated image frames may contain any suitable image-related data, such as RGB image data, YUV image data, or raw image data.
  • FIGURES 10A, 10B, 11A, and 11B illustrate other example imaging arrays 1100 and 1150 for use with array-based depth estimation and related details
  • FIGURES 10A, 10B, 11A, and 11B illustrate other example imaging arrays 1100 and 1150 for use with array-based depth estimation and related details
  • these figures are merely meant to illustrate examples of possible alternative arrangements of imaging sensors within imaging arrays.
  • any number of imaging sensors may be used in any suitable arrangement, as long as the imaging sensors define multiple different baseline directions between various ones of the imaging sensors.
  • imaging sensors 202, 1002, 1102, and 1152 have been described as being used to capture reference image frames, while other sensors 204-206, 1004-1006, 1104-1108, and 1154-1162 have been described as being used to capture non-reference image frames.
  • the specific selection of the imaging sensor used to capture reference image frames can vary based on the implementation. In fact, the specific selection of the imaging sensor used to capture reference image frames can vary dynamically if desired.
  • FIGURE 12 illustrates an example method 1200 for array-based depth estimation in accordance with this disclosure.
  • the method 1200 of FIGURE 12 may be described as being performed by the electronic device 101 of FIGURE 1, which may use image frames captured using the imaging array 200 of FIGURE 2.
  • the method 1200 may be performed using any suitable device(s) having any suitable imaging array(s) and in any other suitable system(s).
  • At least three input image frames of a scene are obtained at step 1202.
  • This may include, for example, the processor 120 receiving image data from the imaging sensors 202, 204, and 206 and generating image frames 402, 404, and 406 based on the image data.
  • the imaging sensors 202, 204, and 206 themselves may generate image frames 402, 404, and 406 and provide the image frames to the processor 120.
  • the image frames include a reference image frame (such as the image frame 402) and a plurality of non-reference image frames (such as the image frames 404 and 406).
  • Feature maps for the input image frames are identified at step 1204. This may include, for example, the processor 120 using the feature extractors 408, 410, and 412 to generate feature maps 414, 416, and 418 for the input image frames 402, 404, and 406.
  • Cross-correlations are performed between the feature map of the reference image frame and the feature maps of the plurality of non-reference image frames using sliding windows at step 1206.
  • This may include, for example, the processor 120 performing the cross-correlation functions 420 and 422 with sliding windows 604 and 610 that move in different directions. The different directions are based on the baseline directions defined between the imaging sensor used to capture the reference image frame 402 (such as the imaging sensor 202) and the imaging sensors used to capture the non-reference image frames 404 and 406 (such as the imaging sensors 204 and 206).
  • This may also include the cross-correlation functions 420 and 422 producing correlated feature maps 424 and 426 based on the cross-correlations.
  • Disparity maps and confidence maps are generated by the results of the cross-correlations at step 1208.
  • This may include, for example, the processor 120 performing the disparity refinement functions 428 and 430 to convert the correlated feature maps 424 and 426 into disparity maps 432 and 434.
  • This may also include the cross-correlation functions 420 and 422 performing the operations 702 using the correlated feature maps 424 and 426 to produce initial confidence maps 704 and performing the upsampling operations 706 to produce the higher-resolution confidence maps 436 and 438.
  • the disparity maps and confidence maps are fused to produce a final depth map of the scene at step 1210.
  • This may include, for example, the processor 120 performing the fused depth map generation function 440 to scale disparity values contained in the disparity maps 432 and 434 or to scale depth values that are based on the disparity values contained in the disparity maps 432 and 434 using the confidence values contained in the confidence maps 436 and 438.
  • This may also include the fused depth map generation function 440 using the scaled disparity or depth values to identify final depth values contained in the depth map 442.

Abstract

L'invention concerne un procédé qui consiste à obtenir une pluralité de trames d'image, la pluralité de trames d'image comprenant une trame d'image de référence et une pluralité de trames d'image non de référence; à générer une première carte de disparité sur la base de la trame d'image de référence et d'une première trame d'image non de référence; à générer une seconde carte de disparité sur la base de la trame d'image de référence et d'une seconde trame d'image non de référence; à générer, sur la base de la première carte de disparité, une première carte de confiance comprenant une première pluralité de niveaux de confiance associés à une première pluralité de valeurs de disparité de la première carte de disparité; à générer, sur la base de la seconde carte de disparité, une seconde carte de confiance comprenant une seconde pluralité de niveaux de confiance associés à une seconde pluralité de valeurs de disparité de la seconde carte de disparité; et à générer une carte de profondeur de la scène sur la base de la première carte de disparité, de la seconde carte de disparité, de la première carte de confiance et de la seconde carte de confiance.
PCT/KR2021/095070 2020-07-27 2021-06-01 Estimation de profondeur basée sur un réseau WO2022025741A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US202063056999P 2020-07-27 2020-07-27
US63/056,999 2020-07-27
US17/027,106 US11816855B2 (en) 2020-02-11 2020-09-21 Array-based depth estimation
US17/027,106 2020-09-21

Publications (1)

Publication Number Publication Date
WO2022025741A1 true WO2022025741A1 (fr) 2022-02-03

Family

ID=80035955

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2021/095070 WO2022025741A1 (fr) 2020-07-27 2021-06-01 Estimation de profondeur basée sur un réseau

Country Status (1)

Country Link
WO (1) WO2022025741A1 (fr)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120321172A1 (en) * 2010-02-26 2012-12-20 Jachalsky Joern Confidence map, method for generating the same and method for refining a disparity map
US20130129194A1 (en) * 2011-11-21 2013-05-23 Robo-team Ltd. Methods and systems of merging depth data from a plurality of disparity maps
US20140321712A1 (en) * 2012-08-21 2014-10-30 Pelican Imaging Corporation Systems and Methods for Performing Depth Estimation using Image Data from Multiple Spectral Channels
US20150071524A1 (en) * 2013-09-11 2015-03-12 Motorola Mobility Llc 3D Feature Descriptors with Camera Pose Information
KR20160083905A (ko) * 2014-01-03 2016-07-12 인텔 코포레이션 깊이 카메라를 사용하는 실시간 3d 재구성

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120321172A1 (en) * 2010-02-26 2012-12-20 Jachalsky Joern Confidence map, method for generating the same and method for refining a disparity map
US20130129194A1 (en) * 2011-11-21 2013-05-23 Robo-team Ltd. Methods and systems of merging depth data from a plurality of disparity maps
US20140321712A1 (en) * 2012-08-21 2014-10-30 Pelican Imaging Corporation Systems and Methods for Performing Depth Estimation using Image Data from Multiple Spectral Channels
US20150071524A1 (en) * 2013-09-11 2015-03-12 Motorola Mobility Llc 3D Feature Descriptors with Camera Pose Information
KR20160083905A (ko) * 2014-01-03 2016-07-12 인텔 코포레이션 깊이 카메라를 사용하는 실시간 3d 재구성

Similar Documents

Publication Publication Date Title
WO2021201438A1 (fr) Système et procédé de déformation de mouvement utilisant des trames à multiples expositions
WO2019216632A1 (fr) Dispositif électronique et procédé de stockage et de traitement de domaine fovéal
WO2021177784A1 (fr) Génération de carte de profondeur à super-résolution pour caméras multiples ou autres environnements
WO2022019437A1 (fr) Fusion d'images à expositions multiples guidée
WO2020017890A1 (fr) Système et procédé d'association 3d d'objets détectés
WO2021101097A1 (fr) Architecture de réseau neuronal de fusion multi-tâches
WO2021112382A1 (fr) Appareil et procédé de rectification multi-caméra dynamique à l'aide de caméras de profondeur
WO2022014790A1 (fr) Mise à jour de gradient de rétropropagation guidée pour tâche de traitement d'image utilisant des informations redondantes de l'image
WO2022146023A1 (fr) Système et procédé de rendu d'effet de profondeur de champ synthétique pour des vidéos
WO2021162201A1 (fr) Interface utilisateur d'appareil photographique à zoom à encliquetage
WO2022025565A1 (fr) Système et procédé de génération d'une image de bokeh pour un rendu de profondeur de champ de qualité dslr et procédé d'affinage et d'apprentissage associé
WO2021256812A1 (fr) Moteur de fusion d'images synthétiques multi-capteurs, multi-vues, multi-trames et multi-tâches pour système d'imagerie mobile
US11816855B2 (en) Array-based depth estimation
WO2022025741A1 (fr) Estimation de profondeur basée sur un réseau
WO2023043116A1 (fr) Segmentation sémantique sensible à la qualité d'image destinée à être utilisée dans l'entraînement de réseaux antagonistes de génération d'image
WO2021107592A1 (fr) Système et procédé de retouche d'image précise pour éliminer un contenu non souhaité d'images numériques
WO2021025375A1 (fr) Appareil et procédé d'alignement d'image régularisé efficace pour fusion multi-trame
WO2021101037A1 (fr) Système et procédé de sélection dynamique de trame d'image de référence
WO2021221492A1 (fr) Systèmes et procédés d'évaluation quantitative de qualité de carte optique et d'automatisation d'augmentation de données
WO2024071612A1 (fr) Dispositif de réalité augmentée (ar) à vidéo en transparence (vst) et son procédé de fonctionnement
WO2022197066A1 (fr) Mélange de pixels pour synthétiser des trames vidéo avec gestion d'occlusion et de tatouage numérique
WO2023149786A1 (fr) Procédé et dispositif électronique de synthèse de données d'apprentissage d'image et de traitement d'image à l'aide d'une intelligence artificielle
WO2021096152A1 (fr) Couche de corrélation normalisée asymétrique pour adaptation de caractéristiques de réseau neuronal profond
WO2023146329A1 (fr) Procédé et dispositif électronique de non-distorsion faciale dans des images numériques à l'aide de multiples capteurs d'imagerie
WO2024048886A1 (fr) Techniques de fusion pour combiner des bits les plus significatifs et des bits les moins significatifs de données d'image dans un traitement d'image ou d'autres applications

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21848700

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21848700

Country of ref document: EP

Kind code of ref document: A1