WO2019061713A1 - 一种基于机器人的目标图像显示处理方法及系统 - Google Patents

一种基于机器人的目标图像显示处理方法及系统 Download PDF

Info

Publication number
WO2019061713A1
WO2019061713A1 PCT/CN2017/110287 CN2017110287W WO2019061713A1 WO 2019061713 A1 WO2019061713 A1 WO 2019061713A1 CN 2017110287 W CN2017110287 W CN 2017110287W WO 2019061713 A1 WO2019061713 A1 WO 2019061713A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
frame
kth
feature point
target
Prior art date
Application number
PCT/CN2017/110287
Other languages
English (en)
French (fr)
Inventor
谢阳阳
郭昌野
Original Assignee
南京阿凡达机器人科技有限公司
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
Application filed by 南京阿凡达机器人科技有限公司 filed Critical 南京阿凡达机器人科技有限公司
Priority to US15/972,073 priority Critical patent/US10733739B2/en
Publication of WO2019061713A1 publication Critical patent/WO2019061713A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle

Definitions

  • the invention relates to the field of robots, in particular to a robot-based target image display processing method and system.
  • the human body follows the advantages of human-computer interaction, has the advantages of convenience and high degree of automation, and is widely used in robots with walking ability. Specifically, when the robot implements the human body following technology, the image signal is first collected by a built-in camera and the like, and when the human body target is detected in the collected image signal, the human body image is displayed on the display screen of the robot and marked. The position of the human body moves with the movement of the human body.
  • the acquisition speed of the image by the robot is much higher than the processing speed of the image.
  • the acquisition module has acquired the image of the nth frame.
  • the system will display the detection result to the position of n+1 frame on the time axis.
  • the detection module skips the detection of the 2nd to n-1 frames and starts detecting the content of the nth frame.
  • the acquisition module has already collected the 2n-1th frame, and the system The detection result shows the position of 2n frames on the time axis.
  • the display module between n+1 and 2n-1 frames on the time axis always displays the detection result of the first frame, and the detection result of the nth frame is displayed between 2n and 3n-2 frames on the time axis. It can be seen that although the target moves continuously in the field of view, and the display module cannot follow the target in time, n-1 identical frames will be continuously displayed and then jumped to the next nth frame, so on the time axis There is a jam between n+1 frames and 2n-1 frames (the first frame is always displayed), and there is a jump between 2n-1 frames and 2n frames (from displaying the first frame It will become the nth frame), which will bring a bad user experience.
  • An object of the present invention is to provide a robot-based target image display processing method and system, which can solve the problem that the robot has a stuck and jump when displaying a target image, so that the display is smoother and the user experience is improved, and the technical solution is as follows: :
  • a robot-based target image display processing method includes the steps of: S10 continuously collecting video frame images; S20 detecting the following target in the kth frame when detecting a following target in the acquired k-th frame image Position information of the picture; S30 displays the kth frame picture at the k+N+1 frame picture of the time axis, and marks the position of the following target in the kth frame picture; N is a detection period The number of collected frame pictures; S40 sequentially predicts the position of the following target in the acquired k+N+1 to k+2N-1 frame pictures according to the position information of the following target in the kth frame picture; S50 sequentially displays the k+N+1 to k+2N-1 frame pictures and the position of the predicted following target at the k+N+2 to k+2N frame pictures of the time axis.
  • the position information in the position can also avoid the phenomenon that the detected target position changes too much and the picture jumps, so that the displayed image containing the following target is more consistent.
  • step S40 according to the position information of the following target in the kth frame picture, sequentially predicting the position of the following target in the acquired k+N+1 to k+2N-1 frame pictures is specifically Predicting the position of the following target in the k+N+1 frame according to the position information of the following target in the kth frame picture, and predicting k+ according to the position of the following target in the k+N+i frame The position following the target in the N+i+1 frame; where i is a positive integer of 1 to N-2, respectively.
  • the position information of the k+N+1 frame picture can be predicted by the position information of the kth frame picture, and the k+N+2 is predicted by the position information of the k+N+1 frame picture.
  • the position information of the frame picture and then through the recursive idea, can predict the position of the following target in the k+N+1 to k+2N-1 frame picture. By predicting, it is possible to avoid the position of the kth frame after the detection and the position of the target image from the position of the k+N+2th frame to the k+2Nth frame on the time axis, so that the displayed image is smoother and does not appear to be stuck. The phenomenon.
  • the method further comprises: detecting the acquired k+N frame picture while displaying the kth frame picture in step S30, and k+2N+1 in the time axis after step S50; The k+N frame picture and the detected position of the following target are displayed at the frame picture.
  • the predicted speed is faster than the detected speed
  • the captured k+N frame picture is detected, and the k+2N+1 frame picture in the time axis is detected.
  • the k+N frame picture and the detected position of the following target are displayed, so that the deviation of the prediction is corrected in time so that the displayed image conforms to the motion trajectory following the target.
  • predicting a position of following the target in the k+N+1 frame is specifically: S41 respectively calculating the corresponding picture of the kth frame following the target The kth feature point information and the acquired k+N+1 frame picture follow the k+N+1 feature point information corresponding to the target; S42 pairs the kth feature point information and the k+N+1 feature Point information is matched to obtain a kth feature point set and a k+N+1 feature point set respectively; S43 calculates a kth centroid information of the kth feature point set according to the kth feature point set, and according to the Calculating a k+N+1 centroid information of the k+N+1 feature point set in the k+N+1 feature point set; S44, according to the kth centroid information, the k+N+1 The centroid information and the central position information of the k-th frame target picture information predict the k+1th frame target predicted picture information.
  • the position of the following target in the picture can be accurately predicted. Achieve the purpose of the forecast.
  • the step of matching the kth feature point information and the k+N+1 feature point information in S42 specifically includes: S421 calculating the kth feature point information according to a K proximity consistency algorithm, and the The similarity of the k+N+1 feature point information is obtained, and the corresponding matching point is obtained; S422 filters out the error matching point according to the Ransac random sampling consensus algorithm to obtain the kth feature point set. And the k+N+1 feature point set.
  • the feature points of each picture can be calculated, and the error matching points are filtered according to the Ransac random sampling consensus algorithm, so that the final feature points are more effective, and the calculation at the time of prediction is more accurate and predictable.
  • the resulting follow target position is more accurate.
  • calculating the k+N+1 centroid information of the k+N+1 feature point set in step S43 specifically includes: S431 calculating, according to the kth feature point set, all features in the kth frame target picture information. Pixel coordinates of the point, and calculating pixel coordinates of all feature points in the k+N+1 frame picture information according to the k+N+1 feature point set; S432 according to all features in the kth frame target picture information The pixel coordinate of the point is calculated to obtain the kth centroid pixel coordinate as the kth centroid information, and the k+N+1 pixel is calculated according to pixel coordinates of all feature points in the k+N+1 frame picture information. The coordinate information is used as the k+N+1 centroid information.
  • P(k)(j) represents the pixel coordinate of the jth feature point in the kth feature point set corresponding to the kth frame target picture information
  • C(k) represents the kth centroid pixel coordinate
  • P(k+N+1)(j) represents the pixel coordinate of the jth feature point in the k+N+1 feature point set corresponding to the k+N+1 frame picture information
  • C(k+N+1) ) represents the k+N+1 centroid pixel coordinates.
  • the centroid coordinate of the following target can be calculated, so that the predicted position of the following target in the picture is more accurate.
  • the position of the following target in the k+N+1th frame picture predicted in step S44 is specifically: S441 calculates the central position pixel coordinate of the kth frame target picture information as the central position information; S442 according to the k centroid pixel coordinates, the k+N+1 centroid image
  • the pixel coordinates of the center position of the k-th frame image information are calculated by the prime coordinates and the central position pixel coordinates of the k-th frame image information, and the calculation formula is as follows:
  • R(k+N+1) R(k)-C(k)+C(k+N+1)
  • R(k+N+1) represents the center position pixel coordinate of the k+N+1 frame picture information
  • R(k) represents the center position pixel coordinate of the kth frame target picture information
  • C(k) ) represents the kth centroid pixel coordinate
  • C(k+N+1) the k+N+1 centroid pixel coordinate
  • a robot-based target image display processing system includes: an acquisition module for continuously acquiring a video frame image; and a detection module electrically connected with the acquisition module, and detecting a follow target when the captured k-th frame image is detected, Detecting location information of the following target in the kth frame picture; a display module electrically connected to the detection module, configured to display the kth frame picture at a k+N+1 frame picture of a time axis, and Marking the position of the following target in the kth frame picture; N is the number of frame pictures collected in one detection period; and the prediction module is electrically connected to the detection module for using the following target according to the following target
  • the position information of the kth frame picture sequentially predicts the position of the following target in the acquired k+N+1 to k+2N-1 frame pictures; the display module is also used for the k+N in the time axis
  • the k+N+1 to k+2N-1 frame pictures and the positions of the predicted following targets are sequentially displayed at +2
  • the position information in the position can also avoid the phenomenon that the detected target position changes too much and the picture jumps, so that the displayed image containing the following target is more consistent.
  • the prediction module is further configured to predict, according to the location information of the following target in the kth frame picture, a position of the following target in the k+N+1 frame, and according to the k+N+i frame.
  • the position following the target is predicted, and the position following the target in the k+N+i+1 frame is predicted; where i is a positive integer of 1 to N-2, respectively.
  • the k+N+1 can be predicted by the position information of the kth frame picture.
  • the position information of the frame picture is predicted by the position information of the k+N+1 frame picture to obtain the position information of the k+N+2 frame picture, and then through the recursive idea, the k+N+1 to kth can be predicted.
  • the position of the target in the +2N-1 frame picture By predicting, it is possible to avoid the position of the kth frame after the detection and the position of the target image from the position of the k+N+2th frame to the k+2Nth frame on the time axis, so that the displayed image is smoother and does not appear to be stuck. The phenomenon.
  • the detecting module is further configured to: when displaying the kth frame picture, detecting the collected k+N frame picture; the display module is further configured to use the k+2N+1 frame in the time axis.
  • the k+N frame picture and the detected position of the following target are displayed at the picture.
  • the predicted speed is faster than the detected speed
  • the captured k+N frame picture is detected, and the k+2N+1 frame picture in the time axis is detected.
  • the k+N frame picture and the detected position of the following target are displayed, so that the deviation of the prediction is corrected in time so that the displayed image conforms to the motion trajectory following the target.
  • the prediction module further includes: a calculation submodule, configured to respectively calculate a kth feature point information corresponding to the kth frame following object and a corresponding k+N+1 frame picture following target corresponding to the target a k+N+1 feature point information; a matching submodule electrically connected to the calculation submodule, and matching the kth feature point information and the k+N+1 feature point information to obtain a kth a set of feature points, a set of k+N+1 feature points; the calculation sub-module, further configured to calculate a k-th centroid information of the k-th feature point set according to the k-th feature point set, and according to the Calculating a k+N+1 centroid information of the k+N+1 feature point set by the k+N+1 feature point set; and a prediction submodule, configured to use the kth centroid information, the kth The +N+1 centroid information and the central position information of the k-th frame target picture information predict the
  • the position of the following target in the picture can be accurately predicted. Achieve the purpose of the forecast.
  • the matching sub-module further includes: a picture processing unit, configured to calculate a similarity between the k-th feature point information and the k+N+1 feature point information according to a K-proximity consistency algorithm, to obtain a corresponding Matching point; the picture processing unit is also used for random selection according to Ransac
  • the coincidence algorithm filters out the error matching points to obtain the kth feature point set and the k+N+1 feature point set.
  • the feature points of each picture can be calculated, and the error matching points are filtered according to the Ransac random sampling consensus algorithm, so that the final feature points are more effective, and the calculation at the time of prediction is more accurate and predictable.
  • the resulting follow target position is more accurate.
  • the calculation sub-module is further configured to calculate, according to the k-th feature point set, pixel coordinates of all feature points in the k-th frame target picture information, and calculate according to the k+N+1 feature point set Obtaining pixel coordinates of all feature points in the k+N+1 frame picture information; the calculation submodule is further configured to calculate the kth centroid according to pixel coordinates of all feature points in the kth frame target picture information The pixel coordinates are used as the kth centroid information, and the k+N pixel coordinate information is calculated as the k+N+1 centroid information according to the pixel coordinates of all the feature points in the k+N+1 frame picture information.
  • the calculation submodule is further configured to calculate the coordinates of the kth centroid pixel, and the calculation formula is as follows:
  • P(k)(j) represents the pixel coordinates of the jth feature point in the kth feature point set corresponding to the target picture of the kth frame
  • C(k) represents the kth centroid pixel coordinate
  • the calculation sub-module is further configured to calculate the coordinates of the k+N+1 centroid pixel, and the calculation formula is as follows;
  • P(k+N+1)(j) represents the pixel coordinate of the jth feature point in the k+N+1 feature point set corresponding to the k+N+1 frame picture information
  • C(k+N+1) ) represents the k+N+1 centroid pixel coordinates.
  • the calculation sub-module is further configured to calculate a central location pixel coordinate of the k-th frame target image information as central location information;
  • the calculation submodule is further configured to calculate the k+ according to the kth centroid pixel coordinate, the k+N+1 centroid pixel coordinate, and the central position pixel coordinate of the kth frame image information.
  • the center position pixel coordinate of the N+1 frame picture information is calculated as follows:
  • R(k+N+1) R(k)-C(k)+C(k+N+1)
  • R(k+N+1) represents the center position pixel coordinate of the k+N+1 frame picture information
  • R(k) represents the center position pixel coordinate of the kth frame target picture information
  • C(k) ) represents the kth centroid pixel coordinate
  • C(k+N+1) the k+N+1 centroid pixel coordinate
  • the detection speed of the robot is far less than the speed of the acquisition, and when the robot displays the picture following the target, the phenomenon of jamming or image jumping often occurs, which is provided by the present invention.
  • the prediction function can predict the display picture during the detection, so that the displayed picture is smooth and avoids the phenomenon of stagnation.
  • FIG. 1 is a flow chart of an embodiment of a robot-based target image display processing method according to the present invention
  • FIG. 2 is a flow chart of another embodiment of a robot-based target image display processing method according to the present invention.
  • FIG. 3 is a flow chart of another embodiment of a robot-based target image display processing method according to the present invention.
  • FIG. 5 is a schematic structural diagram of a robot-based target image display processing system according to the present invention.
  • FIG. 6 is another schematic structural diagram of a robot-based target image display processing system according to the present invention.
  • FIG. 7 is another schematic structural diagram of a robot-based target image display processing system according to the present invention.
  • FIG. 9 is a timing chart of a robot-based target image display processing method according to the present invention.
  • FIG. 10 is a flow chart of a prediction step in the embodiment of the robot-based target image display processing method corresponding to FIG. 4;
  • FIG. 11 is an image matching relationship diagram in the embodiment of the robot-based target image display processing method corresponding to FIG. 4.
  • FIG. 11 is an image matching relationship diagram in the embodiment of the robot-based target image display processing method corresponding to FIG. 4.
  • 1- acquisition module 2-detection module, 3-prediction module, 4-display module, 31-calculation sub-module, 32-match sub-module, 33-prediction sub-module, 321-picture processing unit.
  • the present invention provides an embodiment of a robot-based target image display processing method, including the steps of:
  • S30 displays the kth frame picture at the k+N+1 frame of the time axis, and marks the position of the following target in the kth frame picture; N is the number of pictures collected in one detection period;
  • S40 sequentially predict, according to the position information of the following target in the kth frame picture, the position of the following target in the acquired k+N+1 to k+2N-1 frame pictures;
  • S50 sequentially displays the k+N+1 to k+2N-1 frame pictures and the position of the predicted following target at the k+N+2 to k+2N frames of the time axis.
  • This embodiment can be applied to a machine having a function of following a target, such as a robot having a human body following function.
  • the k value is taken as 1, that is, the detection is started from the kth frame picture. Since the performance of different robots is different, the number of picture information collected in one detection cycle is also different. The value of the number of pictures N collected during one detection period will change, that is, N can be a detection. The number of images collected during the cycle.
  • the collected picture is referred to as picture information
  • the picture that is displayed after being processed is referred to as target picture information.
  • a camera is installed on the robot to collect an image that follows the target.
  • the robot starts to continuously collect the image containing the following target; when the first frame image information is collected, The picture information of the first frame will be detected. Since the detection speed is far less than the speed of the picture collection, when the picture information of the first frame is detected, the robot has already collected the picture of the N+2 frame. At this time, the robot The first frame picture information that has been detected is displayed at the position of the N+2th frame on the time axis.
  • the conventional technique is to continue to detect the collected nth frame picture, and then display it at the 2nth, the picture of the n+1th frame to the 2n-1th frame on the display is always Keep the picture of the first frame, this will cause the display to appear stuck, and by the 2n frame When a new picture suddenly appears, an image jump will occur, which greatly affects the user's visual experience.
  • the present embodiment provides a prediction function. Since the robot continuously collects picture information, after detecting the first frame picture information and displaying it, step S30 ends, step S40 is started; and the first frame target is obtained according to the detection.
  • the picture information and the position of the following target and the collected N+1 frame picture information predict the position of the second frame target picture information and the following target, and display the predicted second position at the position of the N+2 frame of the time axis.
  • Frame target picture information and the location of the target By repeating the above prediction function, the third frame target picture information can be obtained until the 2N-1 frame target predicted picture information and the position of the following target are predicted and displayed.
  • the k-th frame containing the following target image is acquired, the position of the following target image in the k-th frame picture is detected, and after the detection is completed, the k-th frame is displayed at the k+N+1 frame of the time axis.
  • the prediction function provided in this embodiment can compensate for the phenomenon of frequency hopping and jamming in the prior art.
  • the period required for detecting each frame is N, and the following target is detected from the kth frame; then the kth frame is detected at the k+N frame position, and the kth is displayed at the k+N+1 position.
  • the position of the following target in the k+N+2 frame is predicted according to the position of the target in the predicted k+N+1 frame until the position of the following target in the k+2N-1 frame is predicted; Just the end of the k+N frame detection, the k+N frame picture and the position of the following target are displayed at the k+2N+1 position.
  • the present invention provides another embodiment of a robot-based target image display processing method, including the steps of:
  • S30 displays the kth frame picture at the k+N+1 frame of the time axis, and marks the following target Marked in the position of the kth frame picture; N is the number of pictures collected during one detection period;
  • S44 predicts, according to the kth centroid information, the k+N+1 centroid information, and the center position information of the kth frame target picture information, the position of the following target in the k+N+1 frame picture.
  • S50 sequentially displays the k+N+1 to k+2N-1 frame pictures and the position of the predicted following target at the k+N+2 to k+2N frames of the time axis.
  • how to predict the position of the following target in the acquired k+N+1 frame picture is specifically described based on the detected target picture information of the kth frame and the position of the following target.
  • the k+N+1 feature point information corresponding to the position of the target in the middle such as the pixel coordinates of each feature point in the picture
  • the kth feature point information and the k+N+1 feature point information Performing matching, respectively obtaining a kth feature point set and a k+N+1 feature point set, wherein the feature points in the two feature point sets are two-to-two correspondence; again, according to the k-th feature point set,
  • the k+N+1 feature point set is respectively calculated, and the kth centroid information of the kth feature point set and the k+N+1 centroid
  • the present invention provides another embodiment of a robot-based target image display processing method, including the steps of:
  • S30 displays the kth frame picture at the k+N+1 frame of the time axis, and marks the position of the following target in the kth frame picture; N is the number of pictures collected in one detection period;
  • S421 calculates a similarity between the kth feature point information and the k+N+1 feature point information according to a K proximity consistency algorithm, to obtain a corresponding matching point;
  • S422 filters out the error matching points according to the Ransac random sampling consensus algorithm, and obtains the kth feature point set and the k+N+1 feature point set.
  • S431 calculating pixel coordinates of all feature points in the target picture information of the kth frame according to the kth feature point set, and calculating the k+N+1 frame picture information according to the k+N+1 feature point set The pixel coordinates of all feature points;
  • S432 calculating, according to pixel coordinates of all feature points in the target picture information of the kth frame, the kth centroid pixel coordinates as the kth centroid information, and according to all the feature points in the k+N+1 frame picture information.
  • the pixel coordinates are calculated to obtain the k+N+1 pixel coordinate information as the k+N+1 centroid information.
  • S44 predicts, according to the kth centroid information, the k+N+1 centroid information, and the center position information of the kth frame target picture information, the position of the following target in the k+N+1 frame picture.
  • S50 sequentially displays the k+N+1 to k+2N-1 frame pictures and the position of the predicted following target at the k+N+2 to k+2N frames of the time axis.
  • this embodiment mainly describes how to match the kth feature point information and the k+N+1 feature point information to obtain a kth feature point set, a process of the k+N+1 feature point set; and how to calculate the kth centroid information of the kth feature point set according to the kth feature point set and the k+N+1 feature point set respectively
  • the process of the k+N+1 centroid information of the k+N+1 feature point set :
  • the kth feature point information corresponding to the kth frame target picture information and the k+N feature point information corresponding to the acquired k+N frame picture information may be separately calculated by the feature point detection method, such as using FAST
  • the feature point detection method obtains the kth feature point information corresponding to the kth frame target picture information and the k+N+1 feature point information corresponding to the acquired k+N+1 frame picture information;
  • the similarity between the kth feature point information and the k+N+1 feature point information is calculated according to the K proximity consistency algorithm to obtain a corresponding matching point.
  • the two pictures are different, and the calculated feature points are not all the same. Therefore, it is necessary to calculate the kth feature point information and the k+N+1 feature point information according to the K proximity consistency algorithm. Similarity, when two corresponding feature points reach a certain degree of similarity, it is determined that the two corresponding feature points are valid feature points, thereby obtaining matching points corresponding to the two pictures.
  • the target picture in the kth frame has three feature points A1, B1, and C1
  • the k+N+1 frame picture has three feature points A2, B2, and D2, and A1 and A2, B1 are obtained by the K-proximity consistency algorithm.
  • the feature point similarity with B2 satisfies the judgment requirement, so it can be determined as the corresponding matching point, and the feature point similarity of C1 and D2 does not meet the judgment requirement, then the two feature points are judged not to be the corresponding matching point;
  • the error matching points are filtered according to the Ransac random sampling consensus algorithm to obtain the kth feature point set and the k+N+1 feature point set.
  • the feature points calculated by the image are not only the feature points of the target, but also some feature points in other background images. These feature points are not the feature points needed for detection and prediction, so they need to be randomly based on Ransac.
  • the sampling consensus algorithm filters out the error matching points to obtain correct feature points, and obtains the kth feature point set and the k+N+1 feature point set.
  • pixel coordinates of all feature points in the k-th frame target picture information and all the k+N+1 frame picture information are calculated.
  • the pixel coordinates of the feature point; the pixel coordinates may establish a coordinate system with one corner of the image as a vertex, and then obtain the pixel coordinates corresponding to each feature point by calculating the position of each feature point in the coordinate system.
  • the present invention provides another embodiment of a robot-based target image display processing method, including the steps of:
  • S30 displays the kth frame picture at the k+N+1 frame of the time axis, and marks the position of the following target in the kth frame picture; N is the number of pictures collected in one detection period;
  • S421 calculates a similarity between the kth feature point information and the k+N+1 feature point information according to a K proximity consistency algorithm, to obtain a corresponding matching point;
  • S422 filters out the error matching points according to the Ransac random sampling consensus algorithm, and obtains the kth feature point set and the k+N+1 feature point set.
  • S431 calculating pixel coordinates of all feature points in the target picture information of the kth frame according to the kth feature point set, and calculating the k+N+1 frame picture information according to the k+N+1 feature point set The pixel coordinates of all feature points;
  • S432 calculating, according to pixel coordinates of all feature points in the target picture information of the kth frame, the kth centroid pixel coordinates as the kth centroid information, and according to all the feature points in the k+N+1 frame picture information. Calculating, by pixel coordinates, the k+N+1 pixel coordinate information as the k+N+1 centroid information;
  • P(k)(j) represents the pixel coordinate of the jth feature point in the kth feature point set corresponding to the kth frame target picture information
  • C(k) represents the kth centroid pixel coordinate
  • P(k+N+1)(j) represents the pixel coordinate of the jth feature point in the k+N+1 feature point set corresponding to the k+N+1 frame picture information, C(k+N+1) ) indicating the k+N+1 centroid pixel coordinates;
  • S441 calculates a central position pixel coordinate of the k-th frame target picture information as central position information
  • S442 calculates a center of the k+N+1 frame picture information according to the kth centroid pixel coordinate, the k+N+1 centroid pixel coordinate, and the center position pixel coordinate of the kth frame picture information.
  • Position pixel coordinates which are calculated as follows:
  • R(k+N+1) R(k)-C(k)+C(k+N+1)
  • R(k+N+1) represents the center position pixel coordinate of the k+N+1 frame picture information
  • R(k) represents the center position pixel coordinate of the kth frame target picture information
  • C(k) ) represents the kth centroid pixel coordinate
  • C(k+N+1) the k+N+1 centroid pixel coordinate
  • S50 sequentially displays the k+N+1 to k+2N-1 frame pictures and the position of the predicted following target at the k+N+2 to k+2N frames of the time axis.
  • the embodiment specifically describes how to calculate the kth centroid pixel coordinate, the k+N+1 pixel coordinate information, and how to according to the kth centroid pixel coordinate, the k+N+1
  • the centroid pixel coordinates and the central position pixel coordinates of the k-th frame picture information are calculated, and the center position pixel coordinates of the k+N+1 frame picture information are calculated.
  • the centroid coordinates of the picture information of the kth frame can be calculated according to the formula of step S433; for example, the kth is calculated.
  • the frame target picture information has three feature points, and their pixel coordinates are (2, 5), (1, 6), (3, 7), respectively, and the centroid coordinate can be calculated by the formula of step S433 (2, 6). ).
  • the centroid coordinate of the k+N+1 frame picture information can also be calculated according to the formula in step S434.
  • the pixel coordinates of the central position of the target picture information of the kth frame are calculated, because in the prediction process, the central coordinate position of the picture information and the centroid coordinate position in the picture always have an approximate relative distance, here for convenience , taking an equal distance, so the formula in step S450 can be derived.
  • step S450 Calculating the center position of the picture information of the k+N+1 frame according to the formula described in step S450 Set the pixel coordinates. According to the pixel coordinates of the center position of the k+N+1 frame picture information and the centroid coordinates of the following target, it can be calculated which position of the following target in the k+N+1 frame picture is specific, thereby predicting the k+N +1 frame picture information. Through the above method, the position of the following target in the acquired k+N+1th to k+2N-1th frame pictures is sequentially predicted.
  • the robot can judge how the following target moves, and adjust its position according to the result of the detection.
  • the result rect 1 is detected by the first frame 1 (ie, the part of the upper body of the frame 1 is included in the frame 1, such as the small rectangular frame on the left side in FIG. 11) and the n+1th frame is estimated.
  • the second frame displayed is an example.
  • the feature points in the rect 1 and the feature points of the entire n+1th frame are calculated, for example, FAST feature points.
  • the specific matching method is as follows: the K-nearest neighbor consistency algorithm is used to calculate the feature point similarity of the two graphs, and the error matching points are filtered out based on the Ransac random sampling consensus algorithm, and finally the inner point set P(1) and frame respectively belonging to frame 1 are obtained.
  • the inner point set P(n+1) of n+1, the two sets of point sets are one-to-one correspondence.
  • the centroids of P(1) and P(n+1) are respectively calculated by the formula in step S432; then: the position of the upper body of the human body in the image of the second frame is predicted; In the frame, the center position of the target key point and the center position of the rectangular frame of the target are always approximate relative distances, and here, for convenience, the equal distance is taken. Therefore, the formula in step S442 can be obtained, and the rectangular frame size of rect n+1 is equal to rect 1. As shown on the right side of FIG. 11, the predicted result is a dotted rectangle rectangle rect n+1, so that the display module can display the collected n+1th frame picture and the predicted dotted rectangle frame rect n+1 as the second frame. image.
  • the acquisition module collects the next frame, repeating the above operations, the position of the predicted upper body target can be continuously displayed.
  • the time-consuming part is to calculate the feature points, and the calculation of the feature points only needs to compare the gray levels of the pixels in the image and the neighborhood 3*3 in turn, and the processing speed is fast (the comparison algorithm is simple);
  • the processing of the detection module a large number of convolution operations are required on the image to extract feature points, and the parameters of the convolution kernel are up to several million, and a large number of floating-point operations are compared. time consuming. Therefore, the processing speed of the prediction module is much higher than the detection speed of the detection module.
  • the present invention provides an embodiment of a robot-based target image display processing system, including:
  • the acquisition module 1 is configured to continuously collect video frame images
  • the detecting module 2 is electrically connected to the collecting module 1 and detects position information of the following target in the kth frame picture when the following target is detected in the acquired kth frame picture;
  • the display module 4 is electrically connected to the detecting module 2, configured to display the kth frame picture at the k+N+1 frame picture of the time axis, and mark the following target in the kth frame picture Position; N is the number of frame pictures collected during one detection period;
  • the prediction module 3 is electrically connected to the detection module, and is configured to sequentially predict the acquired k+N+1th to k+2N-1 frames according to the position information of the following target in the kth frame picture. The position of the image following the target;
  • the display module 4 is further configured to sequentially display the k+N+1 to k+2N-1 frame pictures and the predicted following target at the k+N+2 to k+2N frame pictures of the time axis. position.
  • the acquisition module may be composed of a camera mounted on the robot for collecting an image following the target.
  • the robot starts to continuously collect the image containing the following target;
  • the detection module of the robot just detects the first frame picture information to obtain the first frame target picture information, and displays the detected first frame target picture through the display module at the position of the N+1th frame on the time axis.
  • the robot skips the detection of the picture information of the N+2 frame to the picture information of the 2Nth frame, directly detects the picture information of the (N+1)th frame, and displays the N+th position after the detection of the position of the 2Nth frame of the time axis. 1 frame of target picture information.
  • the robot continuously collects picture information, after detecting the first frame picture information and displaying it, according to the detected first frame target picture information and the collected N+1 frame picture information, the second frame target picture information is predicted. And the predicted second frame target picture information is displayed at the position of the N+2 of the time axis.
  • the target picture information of the third frame can be obtained, until the prediction function of the second N-1 frame target predicted picture information is predicted and displayed, which can compensate for the frequency hopping and the stagnation of the prior art. The phenomenon.
  • the present invention provides another embodiment of a robot-based target image display processing system, including:
  • the acquisition module 1 is configured to continuously collect video frame images
  • the detecting module 2 is electrically connected to the collecting module 1 and detects position information of the following target in the kth frame picture when the following target is detected in the acquired kth frame picture;
  • the display module 4 is electrically connected to the detecting module 2, configured to display the kth frame picture at the k+N+1 frame picture of the time axis, and mark the following target in the kth frame picture Position; N is the number of frame pictures collected during one detection period;
  • the prediction module 3 is electrically connected to the detection module 2, and is configured to sequentially predict the acquired k+N+1 to k+2N-1 according to the position information of the following target in the kth frame picture. The position of the target in the frame picture;
  • the display module 4 is further configured to sequentially display the k+N+1 to k+2N-1 frame pictures and the predicted following target at the k+N+2 to k+2N frame pictures of the time axis. position;
  • the prediction module 3 includes:
  • the calculation sub-module 31 is configured to respectively calculate the k-th feature point information corresponding to the k-th frame picture following target and the k+N+1 feature point information corresponding to the acquired k+N+1-frame picture following target ;
  • the matching sub-module 32 is electrically connected to the calculation sub-module 31, and matches the k-th feature point information and the k+N+1 feature point information to obtain a k-th feature point set and a k+N, respectively. +1 feature point set;
  • the calculation sub-module 31 is further configured to calculate k-th centroid information of the k-th feature point set according to the k-th feature point set, and calculate the (k+N+1-th feature point set according to the k+N+1 feature point set The k+N+1 centroid information of the k+N+1 feature point set;
  • the prediction sub-module 33 is configured to predict, according to the k-th centroid information, the k+N+1 centroid information, and the center position information of the k-th frame target picture information, the k+1th frame target predicted picture information .
  • the image processing module uses the existing processing methods for the feature points in the image to obtain the k-th feature point information corresponding to the k-th frame target picture information and the corresponding k+N+1 frame picture information.
  • the k+N+1 feature point information such as the pixel coordinates of each feature point in the picture; secondly, the kth feature point information and the k+N+1 feature point information are matched, respectively k feature point set, k+N+1 feature point set, the feature points in the two feature point sets are two-to-two correspondence; again, according to the k-th feature point set, the k+N+ a feature point set, wherein the calculation module respectively calculates the kth centroid information of the kth feature point set and the k+N+1 centroid information of the k+N+1 feature point set, wherein the feature information includes a centroid Pixel coordinates in the picture; finally, predicting the k+N+1 frame target based on the kth centroid information, the k+N+1 centroid information, and the center position information of the kth frame target picture information Predict picture information.
  • the present invention provides another embodiment of a robot-based target image display processing system, including:
  • the acquisition module 1 is configured to continuously collect video frame images
  • the detecting module 2 is electrically connected to the collecting module 1 and detects position information of the following target in the kth frame picture when the following target is detected in the acquired kth frame picture;
  • the display module 4 is electrically connected to the detecting module 2, configured to display the kth frame picture at the k+N+1 frame picture of the time axis, and mark the following target in the kth frame picture Position; N is the number of frame pictures collected during one detection period;
  • the prediction module 3 is electrically connected to the detection module 2, and is configured to sequentially predict the acquired k+N+1 to k+2N-1 according to the position information of the following target in the kth frame picture. The position of the target in the frame picture;
  • the display module 4 is further configured to sequentially display the k+N+1 to k+2N-1 frame pictures and the predicted following target at the k+N+2 to k+2N frame pictures of the time axis. position;
  • the prediction module 3 includes:
  • the calculation sub-module 31 is configured to respectively calculate the k-th feature point information corresponding to the k-th frame picture following target and the k+N+1 feature point information corresponding to the acquired k+N+1-frame picture following target ;
  • the matching sub-module 32 is electrically connected to the calculation sub-module 31, and matches the k-th feature point information and the k+N+1 feature point information to obtain a k-th feature point set and a k+N, respectively. +1 feature point set;
  • the calculation sub-module 31 is further configured to calculate k-th centroid information of the k-th feature point set according to the k-th feature point set, and calculate the (k+N+1-th feature point set according to the k+N+1 feature point set The k+N+1 centroid information of the k+N+1 feature point set;
  • the prediction sub-module 33 is configured to predict, according to the k-th centroid information, the k+N+1 centroid information, and the center position information of the k-th frame target picture information, the k+1th frame target predicted picture information ;
  • Matching sub-module 32 includes:
  • the image processing unit 321 is configured to calculate a similarity between the kth feature point information and the k+N+1 feature point information according to a K proximity consistency algorithm, to obtain a corresponding matching point;
  • the image processing unit 321 is further configured to filter out the error matching points according to the Ransac random sampling consistency algorithm, to obtain the kth feature point set and the k+N+1 feature point set;
  • the calculation sub-module 31 is further configured to calculate pixel coordinates of all feature points in the target picture information of the k-th frame according to the k-th feature point set, and calculate the first feature according to the k+N+1 feature point set.
  • the calculating sub-module 31 is further configured to calculate, according to pixel coordinates of all feature points in the target picture information of the k-th frame, the k-th centroid pixel coordinates as the k-th centroid information, and according to the k+N+ The k+th pixel coordinate information is calculated as the k+N+1 centroid information by calculating the pixel coordinates of all the feature points in the 1-frame picture information.
  • the calculation submodule is further configured to calculate the coordinates of the kth centroid pixel, and the calculation formula is as follows:
  • P(k)(j) represents the pixel coordinates of the jth feature point in the kth feature point set corresponding to the target picture of the kth frame
  • C(k) represents the kth centroid pixel coordinate
  • the calculation sub-module is further configured to calculate the coordinates of the k+N+1 centroid pixel, and the calculation formula is as follows;
  • P(k+N+1)(j) represents the pixel coordinate of the jth feature point in the k+N+1 feature point set corresponding to the k+N+1 frame picture information
  • C(k+N+1) ) represents the k+N+1 centroid pixel coordinates.
  • the calculation sub-module 31 is further configured to calculate a central location pixel coordinate of the k-th frame target image information as the central location information;
  • the calculating sub-module 31 is further configured to calculate the kth according to the kth centroid pixel coordinates, the k+N+1 centroid pixel coordinates, and the center position pixel coordinates of the kth frame image information. +N+1 frame image information center position pixel coordinates, the calculation formula is as follows:
  • R(k+N+1) R(k)-C(k)+C(k+N+1)
  • R(k+N+1) represents the center position pixel coordinate of the k+N+1 frame picture information
  • R(k) represents the center position pixel coordinate of the kth frame target picture information
  • C(k) ) represents the kth centroid pixel coordinate
  • C(k+N+1) the k+N+1 centroid pixel coordinate
  • the embodiment specifically describes how to calculate the kth centroid pixel coordinate, the k+N+1 pixel coordinate information, and how to according to the kth centroid pixel coordinate, the k+N+1
  • the centroid pixel coordinates and the central position pixel coordinates of the k-th frame picture information are calculated, and the center position pixel coordinates of the k+N+1 frame picture information are calculated.
  • the embodiment mainly describes how to match the kth feature point information and the k+N+1 feature point information to obtain a kth feature point set and a k+N+1 feature point set respectively. And calculating, according to the kth feature point set and the k+N+1 feature point set, the kth centroid information and the k+N+1 feature point of the kth feature point set respectively The process of the k+N+1 centroid information of the set.
  • the matching sub-module can respectively obtain the k-th feature point information corresponding to the k-th frame target picture information and the k+N+1 feature corresponding to the acquired k+N+1-frame picture information by using the feature point detection method.
  • Point information such as using the FAST feature point detection method to obtain the kth feature point information corresponding to the kth frame target picture information and the k+N+1 feature point information corresponding to the acquired k+N+1 frame picture information;
  • the picture processing module can obtain the corresponding matching point.
  • the two pictures are different, and the calculated feature points are not all the same. Therefore, it is necessary to calculate the kth feature point information and the k+N+1 feature point information according to the K proximity consistency algorithm. Similarity, get the matching points corresponding to the two pictures.
  • the error matching points are filtered according to the Ransac random sampling consensus algorithm to obtain the kth feature point set and the k+N+1 feature point set.
  • the feature points calculated by the image are not only the feature points of the target, but also some feature points in other background images. These feature points are not the feature points needed for detection and prediction, so they need to be randomly based on Ransac.
  • the sampling consensus algorithm filters out the error matching points to obtain the correct feature points, that is, the k-th feature point set and the k+N+1 feature point set.
  • pixel coordinates of all feature points in the k-th frame target picture information and all the k+N+1 frame picture information are calculated.
  • the pixel coordinates of the feature point; the pixel coordinates may establish a coordinate system with one corner of the image as a vertex, and then obtain the pixel coordinates corresponding to each feature point by calculating the position of each feature point in the coordinate system.
  • the centroid coordinates of the picture information of the kth frame may be calculated according to the formula of step S433; for example, the target picture information of the kth frame is calculated.
  • Feature points, their pixel coordinates are (2,5), (1,6), (3,7), and the centroid coordinates can be calculated by formula (2,6).
  • the centroid coordinate of the k+N frame picture information can also be calculated according to the formula in step S434.
  • the central position pixel coordinates of the k+N+1 frame picture information are calculated. According to the pixel coordinates of the center position of the k+N+1 frame picture information, it can be calculated at which position of the k+N+1 frame picture information, thereby predicting the following target in the k+N+1 frame picture. s position.
  • the k+N+2 frame target predicted picture information to the k+2N-1 frame target predicted picture information can also be predicted.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Library & Information Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Analysis (AREA)
  • Manipulator (AREA)
  • Image Processing (AREA)

Abstract

本发明公开了一种基于机器人的目标图像显示处理方法和系统，包括：S10对视频帧图像进行连续采集；S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；S30在时间轴第k+N+1帧处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的图片数量；S40根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；S50在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。通过本发明能够更加流畅地显示图像。

Description

一种基于机器人的目标图像显示处理方法及系统
本申请要求2017年09月29日提交的申请号为：201710911965.X、发明名称为“一种基于机器人的目标图像显示处理方法及系统”的中国专利申请的优先权，其全部内容合并在此。
技术领域
本发明涉及机器人领域，特别是一种基于机器人的目标图像显示处理方法及系统。
背景技术
随着人工智能的逐渐发展，机器人必将融入人们的工作与生活之中，如何提高机器人与人之间的交互智能性，成为了目前研究的热点。
人体跟随作为人机交互的一种方式，具有方便，自动化程度高等优点，被广泛应用在具有行走能力的机器人中。具体的，机器人在实现人体跟随技术时，首先通过内置的摄像机等采集设备采集图像信号，当在采集到的图像信号中检测到人体目标时，在机器人的显示屏幕上显示包含该人体图像并标注人体的位置，并跟随人体的移动而移动。
但是，目前的技术中，机器人对图像的采集速度远远高于对图像的处理速度，如图8所示，假设当检测模块处理完第1帧后，采集模块已经采集完第n帧的图像，系统会把检测结果显示到时间轴上n+1帧的位置。为了避免跟踪丢失目标，检测模块跳过对第2～n-1帧的检测、开始检测第n帧的内容，当处理完第n帧时，采集模块已经采集完第2n-1帧，系统把该检测结果显示在时间轴上2n帧的位置。因此，时间轴上n+1至2n-1帧之间显示模块一直显示的是第一帧的检测结果，时间轴上2n至3n-2帧之间显示的是第n帧的检测结果。从中可以看出，虽然目标在视场内不断的移动，而显示模块并不能及时跟随到目标，将连续显示n-1个相同的帧然后跳到之后的第n帧，因此在时间轴上第n+1帧到2n-1帧之间存在卡顿(始终显示第1帧)，在2n-1帧和2n帧之间存在跳变(从显示第1帧突 然变为显示第n帧)，这样会给用户带来不好的使用体验。
因此，需要设计一种能够能加流畅地显示图像的方法，使显示时能够避免卡顿和跳变，给用户带来更好的体验感。
发明内容
本发明的目的是提供一种基于机器人的目标图像显示处理方法及系统，能够解决机器人显示目标图像时存在卡顿、跳变的问题，使显示更为流畅，提高用户的体验感，技术方案如下：
一种基于机器人的目标图像显示处理方法，包括步骤：S10对视频帧图像进行连续采集；S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；S30在时间轴第k+N+1帧图片处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的帧图片数量；S40根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；S50在时间轴第k+N+2至k+2N帧图片处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
通过本发明，能够在检测跟随目标在图片中位置的同时预测跟随目标在图片中的位置，避免了因为检测速度慢而导致显示出来的跟随目标图像出现卡顿的现象，通过预测跟随目标在图像中的位置信息还能够避免检测出来的目标位置变化过大而出现图片跳变的现象，使显示出来的含有跟随目标的图像更具连贯性。
优选的，步骤S40中根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置具体为：根据所述跟随目标在所述第k帧图片的位置信息，预测在第k+N+1帧中跟随目标的位置，以及根据第k+N+i帧中跟随目标的位置，预测k+N+i+1帧中跟随目标的位置；其中i分别为1～N-2的正整数。
根据递归的方法，能够通过第k帧图片的位置信息预测得到第k+N+1帧图片的位置信息，再通过第k+N+1帧图片的位置信息预测得到第k+N+2 帧图片的位置信息，然后通过递归思想，能够预测出第k+N+1至第k+2N-1帧图片中跟随目标的位置。通过预测能够避免在时间轴上第k+N+2帧到第k+2N帧的位置一直显示检测之后的第k帧图像以及目标图像的位置，使显示的图像更加流畅，不会出现卡顿的现象。
优选的，该方法进一步包括：在步骤S30中显示所述第k帧图片的同时，对采集到的第k+N帧图片进行检测，以及在步骤S50之后，在时间轴第k+2N+1帧图片处显示所述k+N帧图片以及所述检测到的跟随目标的位置。
由于预测的速度快于检测的速度，因此，在预测第k+N+2帧跟随目标位置时，对采集到的第k+N帧图片进行检测，在时间轴第k+2N+1帧图片处显示所述k+N帧图片以及所述检测到的跟随目标的位置，这样及时纠正预测出现的偏差，使得显示的图像符合跟随目标的运动轨迹。
优选的，根据所述跟随目标在所述第k帧图片的位置信息，预测在第k+N+1帧中跟随目标的位置具体为：S41分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；S42对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；S43根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；S44根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+1帧目标预测图片信息。
通过对每一帧图片中跟随目标的特征点的采集，并计算每个有效特征点的位置坐标、图片中心位置的坐标以及跟随目标质心坐标，能够准确地预测出跟随目标在图片中的位置，达到预测的目的。
优选的，S42中对所述第k特征点信息和所述第k+N+1特征点信息进行匹配的步骤具体包括：S421根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；S422根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集 和所述第k+N+1特征点集。
根据K邻近一致性算法能够计算出每张图片的特征点，并根据Ransac随机采样一致算法滤除错误匹配点，使最终得到的特征点的有效程度更高，使预测时的计算更加准确，预测得到的跟随目标位置更准确。
优选的，步骤S43中计算所述第k+N+1特征点集的第k+N+1质心信息具体包括：S431根据所述第k特征点集计算得到第k帧目标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；S432根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N+1像素坐标信息作为第k+N+1质心信息。
所述步骤S432中所述的第k质心像素坐标的计算公式如下：
Figure PCTCN2017110287-appb-000001
其中，P(k)(j)表示第k帧目标图片信息对应的第k特征点集中第j个特征点的像素坐标，C(k)表示所述第k质心像素坐标；
所述步骤S432中所述的第k+N+1质心像素坐标的计算公式如下；
Figure PCTCN2017110287-appb-000002
其中，P(k+N+1)(j)表示第k+N+1帧图片信息对应的第k+N+1特征点集中第j个特征点的像素坐标，C(k+N+1)表示所述第k+N+1质心像素坐标。
通过上述公式，根据检测跟随目标的有效特征点在图片中的像素坐标，可以计算得到跟随目标的质心坐标，使预测得到跟随目标在图片中的位置更为准确。
优选的，步骤S44中预测得到第k+N+1帧图片中跟随目标的位置具体为：S441计算得到所述第k帧目标图片信息的中心位置像素坐标作为中心位置信息；S442根据所述第k质心像素坐标、所述第k+N+1质心像 素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标，其计算公式如下：
R(k+N+1)＝R(k)-C(k)+C(k+N+1)
其中，R(k+N+1)表示所述第k+N+1帧图片信息的中心位置像素坐标，R(k)表示所述第k帧目标图片信息的中心位置像素坐标，C(k)表示所述第k质心像素坐标，C(k+N+1)所述第k+N+1质心像素坐标。
通过跟随目标的质心位置和图片中心坐标位置的相对位置，能够预测出跟随目标在下一帧图片中的具体位置，使预测更加准确。
一种基于机器人的目标图像显示处理系统，包括：采集模块，用于对视频帧图像进行连续采集；检测模块，与采集模块电连接，在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；显示模块，与所述检测模块电连接，用于在时间轴第k+N+1帧图片处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的帧图片数量；预测模块，与所述检测模块电连接，用于根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；所述显示模块，还用于在时间轴第k+N+2至k+2N帧图片处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
通过本发明，能够在检测跟随目标在图片中位置的同时预测跟随目标在图片中的位置，避免了因为检测速度慢而导致显示出来的跟随目标图像出现卡顿的现象，通过预测跟随目标在图像中的位置信息还能够避免检测出来的目标位置变化过大而出现图片跳变的现象，使显示出来的含有跟随目标的图像更具连贯性。
优选的，所述预测模块还用于根据所述跟随目标在所述第k帧图片的位置信息，预测在第k+N+1帧中跟随目标的位置，以及根据第k+N+i帧中跟随目标的位置，预测k+N+i+1帧中跟随目标的位置；其中i分别为1～N-2的正整数。
根据递归的方法，能够通过第k帧图片的位置信息预测得到第k+N+1 帧图片的位置信息，再通过第k+N+1帧图片的位置信息预测得到第k+N+2帧图片的位置信息，然后通过递归思想，能够预测出第k+N+1至第k+2N-1帧图片中跟随目标的位置。通过预测能够避免在时间轴上第k+N+2帧到第k+2N帧的位置一直显示检测之后的第k帧图像以及目标图像的位置，使显示的图像更加流畅，不会出现卡顿的现象。
优选的，所述检测模块还用于显示所述第k帧图片的同时，对采集到的第k+N帧图片进行检测；所述显示模块还用于在时间轴第k+2N+1帧图片处显示所述k+N帧图片以及所述检测到的跟随目标的位置。
由于预测的速度快于检测的速度，因此，在预测第k+N+2帧跟随目标位置时，对采集到的第k+N帧图片进行检测，在时间轴第k+2N+1帧图片处显示所述k+N帧图片以及所述检测到的跟随目标的位置，这样及时纠正预测出现的偏差，使得显示的图像符合跟随目标的运动轨迹。
优选的，所述预测模块还包括：计算子模块，用于分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；匹配子模块，与所述计算子模块电连接，对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；所述计算子模块，还用于根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；预测子模块，用于根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+1帧目标预测图片信息。
通过对每一帧图片中跟随目标的特征点的采集，并计算每个有效特征点的位置坐标、图片中心位置的坐标以及跟随目标质心坐标，能够准确地预测出跟随目标在图片中的位置，达到预测的目的。
优选的，所述匹配子模块还包括：图片处理单元，用于根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；所述图片处理单元，还用于根据Ransac随机采 样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。
根据K邻近一致性算法能够计算出每张图片的特征点，并根据Ransac随机采样一致算法滤除错误匹配点，使最终得到的特征点的有效程度更高，使预测时的计算更加准确，预测得到的跟随目标位置更准确。
优选的，所述计算子模块，还用于根据所述第k特征点集计算得到第k帧目标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；所述计算子模块，还用于根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N像素坐标信息作为第k+N+1质心信息。
所述计算子模块，还用于计算所述第k质心像素坐标，计算公式如下：
Figure PCTCN2017110287-appb-000003
其中，P(k)(j)表示第k帧目标图片对应的第k特征点集中第j个特征点的像素坐标，C(k)表示所述第k质心像素坐标；
所述计算子模块，还用于计算所述第k+N+1质心像素坐标，计算公式如下；
Figure PCTCN2017110287-appb-000004
其中，P(k+N+1)(j)表示第k+N+1帧图片信息对应的第k+N+1特征点集中第j个特征点的像素坐标，C(k+N+1)表示所述第k+N+1质心像素坐标。
通过跟随目标的质心位置和图片中心坐标位置的相对位置，能够预测出跟随目标在下一帧图片中的具体位置，使预测更加准确。
优选的，所述计算子模块，还用于计算得到所述第k帧目标图片信息的中心位置像素坐标作为中心位置信息；
所述计算子模块，还用于根据所述第k质心像素坐标、所述第k+N+1质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标，其计算公式如下：
R(k+N+1)＝R(k)-C(k)+C(k+N+1)
其中，R(k+N+1)表示所述第k+N+1帧图片信息的中心位置像素坐标，R(k)表示所述第k帧目标图片信息的中心位置像素坐标，C(k)表示所述第k质心像素坐标，C(k+N+1)所述第k+N+1质心像素坐标。
通过跟随目标的质心位置和图片中心坐标位置的相对位置，能够预测出跟随目标在下一帧图片中的具体位置，使预测更加准确。
根据本发明提供的一种基于机器人的目标图像显示处理方法和系统，能够带来以下有益效果：
能够解决成跟随系统显示卡顿的情况。在以往技术中，由于受到机器人自身硬件条件的限制，使得机器人的检测速度远不及采集的速度，导致机器人显示跟随目标的图片时，常出现卡顿，或者图像跳变的情况，通过本发明提供的预测功能，能够在检测期间预测得到显示图片，使得显示出来的画面流畅，避免了卡顿的现象。
附图说明
下面将以明确易懂的方式，结合附图说明优选实施方式，对一种基于机器人的目标图像显示处理方法及系统的上述特性、技术特征、优点及其实现方式予以进一步说明。
图1是本发明一种基于机器人的目标图像显示处理方法的实施例流程图；
图2是本发明一种基于机器人的目标图像显示处理方法的另一个实施例流程图；
图3是本发明一种基于机器人的目标图像显示处理方法的另一个实施例流程图；
图4是本发明一种基于机器人的目标图像显示处理方法的另一个实施 例流程图；
图5是本发明一种基于机器人的目标图像显示处理系统的一个结构示意图；
图6是本发明一种基于机器人的目标图像显示处理系统的另一个结构示意图；
图7是本发明一种基于机器人的目标图像显示处理系统的另一个结构示意图；
图8是现有技术中机器人的目标图像显示处理方法的时序流程图；
图9是本发明一种基于机器人的目标图像显示处理方法的时序流程图；
图10是图4对应的基于机器人的目标图像显示处理方法实施例中的预测步骤流程图；
图11是图4对应的基于机器人的目标图像显示处理方法实施例中的图像匹配关系图。
附图标号说明：
1-采集模块、2-检测模块、3-预测模块、4-显示模块、31-计算子模块、32-匹配子模块、33-预测子模块、321-图片处理单元。
具体实施方式
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对照附图说明本发明的具体实施方式。显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图，并获得其他的实施方式。
为使图面简洁，各图中只示意性地表示出了与本发明相关的部分，它们并不代表其作为产品的实际结构。另外，以使图面简洁便于理解，在有些图中具有相同结构或功能的部件，仅示意性地绘示了其中的一个，或仅标出了其中的一个。在本文中，“一个”不仅表示“仅此一个”，也可以表示“多于一个”的情形。
如图1所示，本发明提供了一种基于机器人的目标图像显示处理方法的一个实施例，包括步骤：
S10对视频帧图像进行连续采集；
S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
S30在时间轴第k+N+1帧处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的图片数量；
S40根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；
S50在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
本实施例可以运用在具有跟随目标功能的机器上，比如具有人体跟随功能的机器人。为了方便理解，本实施例中将k值取1，即从第k帧图片开始检测。由于不同机器人的性能有所不同，在一个检测周期内采集到的图片信息的数量也有不同，在一个检测周期内所采集的图片数量N的值会有所变化，也就是说N可以是一个检测周期内所采集的图片数量。为了方便区分，本实施例中，将采集到的图片称为图片信息，将处理过后再显示的图片称为目标图片信息，
在本实施例中，取k＝1，代表从检测到含有跟随目标的第一帧图片开始检测。具体的，在机器人身上会安装有摄像头，用于采集跟随目标的图像，当跟随目标出现在机器人视野范围内时，机器人开始连续采集含有跟随目标的图片；在采集到第1帧图片信息时，将对第1帧图片信息进行检测，由于检测的速度远不及采集图片的速度，因此，当检测完第1帧图片信息的时候，机器人已经采集到了第N+2帧图片了，这时候，机器人会在时间轴上第N+2帧的位置显示已经检测完毕的第1帧图片信息。
如图8所示，通常在此时，以往的技术就是继续检测采集到的第n帧图片，然后在第2n处显示，在显示器上第n+1帧～第2n-1帧的图片就一直保持第1帧的图片，这样就会造成显示器出现卡顿的情况，到第2n帧的时候又 突然出现新的图片，会出现图像跳变的情况，非常影响用户的视觉体验。
因此，本实施例提供了一种预测功能，由于机器人在连续采集图片信息，当检测完第1帧图片信息并显示后，即步骤S30结束，开始执行步骤S40；根据检测得到的第1帧目标图片信息及跟随目标的位置和采集到的第N+1帧图片信息，预测得到第2帧目标图片信息及跟随目标的位置，并在时间轴第N+2帧的位置显示预测出来的第2帧目标图片信息及跟随目标的位置。重复执行上述预测功能，可以得到第3帧目标图片信息，直到预测出并显示出第2N-1帧目标预测图片信息及跟随目标的位置。
如图9所示，在采集到第k帧含有跟随目标的图片时，检测第k帧图片中跟随目标图像的位置，检测完毕之后，在时间轴第k+N+1帧处显示第k帧图片及跟随目标的位置，并根据所述跟随目标在所述第k帧图片的位置信息及跟随目标的位置，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置，并在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置，通过本实施例提供的预测功能，能够弥补现有技术出现跳频和卡顿的现象。
具体的，假设检测处理每一帧需要的周期为N，从第k帧开始检测到跟随目标；则在第k+N帧位置检测完第k帧，在第k+N+1位置显示第k帧；然后在k+N+2～k+2N帧位置依次显示k+N+1～k+2N-1帧的图片，并且根据第k帧中跟随目标的位置预测k+N+1帧中跟随目标的位置，根据预测后k+N+1帧中目标的位置来预测第k+N+2帧中跟随目标的位置，直到预测出第k+2N-1帧中跟随目标的位置；随后正好第k+N帧检测结束，在k+2N+1位置处显示该第k+N帧图片和跟随目标的位置。
如图2所示，本发明提供了一种基于机器人的目标图像显示处理方法的另一个实施例，包括步骤：
S10对视频帧图像进行连续采集；
S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
S30在时间轴第k+N+1帧处显示所述第k帧图片，并标记出所述跟随目 标在所述第k帧图片的位置；N为一个检测周期内所采集的图片数量；
S41分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
S42对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；
S43根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；
S44根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+N+1帧图片中跟随目标的位置。
S50在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
在本实施例中，对如何根据检测得到的第k帧目标图片信息及跟随目标的位置，预测得到在采集到的第k+N+1帧图片中跟随目标的位置进行了具体的阐述。首先，运用现有的一些对图片中特征点的处理方法，分别计算出所述第k帧目标图片中跟随目标的位置对应的第k特征点信息和采集到的第k+N+1帧图片中跟随目标的位置对应的第k+N+1特征点信息，如每个特征点在图片中的像素坐标；其次对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集，这两个特征点集中的特征点都是两两对应的关系；再次，根据所述第k特征点集、所述第k+N+1特征点集，分别计算出所述第k特征点集的第k质心信息、第k+N+1特征点集的第k+N+1质心信息，所述的特征信息包括质心在图片中的像素坐标；最后，根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+N+1帧目标预测图片信息及跟随目标的位置。
如图3所示，本发明提供了一种基于机器人的目标图像显示处理方法的另一个实施例，包括步骤：
S10对视频帧图像进行连续采集；
S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
S30在时间轴第k+N+1帧处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的图片数量；
S41分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
S421根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；
S422根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。
S431根据所述第k特征点集计算得到第k帧目标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；
S432根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N+1像素坐标信息作为第k+N+1质心信息。
S44根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+N+1帧图片中跟随目标的位置。
S50在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
具体的，在上个实施例的基础上，本实施例主要阐述了如何对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集的过程；以及如何根据所述第k特征点集、所述第k+N+1特征点集，分别计算出所述第k特征点集的第k质心信息、第k+N+1特征点集的第k+N+1质心信息的过程：
首先，可以通过特征点检测方法分别计算出所述第k帧目标图片信息对应的第k特征点信息和采集到的第k+N帧图片信息对应的第k+N特征点信息，如用FAST特征点检测方法得到第k帧目标图片信息对应的第k特征点信息和采集到的第k+N+1帧图片信息对应的第k+N+1特征点信息；
其次，根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点。在实际情况中，两张图片不一样，计算得到的特征点也不全是相同的，因此，需要根据K邻近一致性算法计算第k特征点信息和所述第k+N+1特征点信息的相似度，当两个对应的特征点达到一定相似度时，判断此两个对应的特征点为有效特征点，以此得到两张图片对应的匹配点。比如第k帧目标图片里有A1、B1、C1三个特征点，第k+N+1帧图片中有A2、B2、D2三个特征点，通过K邻近一致性算法得到A1和A2、B1和B2的特征点相似度满足判断要求，因此可以判断为对应的匹配点，C1和D2的特征点相似度不符合判断要求，则此两个特征点就判断为不是对应的匹配点；
再次，根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。在实际情况中，图片计算出来的特征点除了跟随目标的特征点以外，还有其他背景图片中的一些特征点，这些特征点其实并不是检测和预测所需要的特征点，所以需要根据Ransac随机采样一致算法滤除错误匹配点，得到正确的特征点，得到所述第k特征点集和所述第k+N+1特征点集。
之后，根据所述第k特征点集、所述第k+N+1特征点集分别计算得到第k帧目标图片信息内所有特征点的像素坐标、第k+N+1帧图片信息内所有特征点的像素坐标；所述的像素坐标可以以图片的一个角为顶点建立坐标系，然后通过计算每个特征点在坐标系中的位置得到每个特征点对应的像素坐标。
最后，根据所述第k帧目标图片信息内所有特征点的像素坐标、所述第k+N+1帧图片信息内所有特征点的像素坐标，计算得到所述第k质心像素坐标、所述第k+N+1像素坐标信息；
如图4所示，本发明提供了一种基于机器人的目标图像显示处理方法的另一个实施例，包括步骤：
S10对视频帧图像进行连续采集；
S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
S30在时间轴第k+N+1帧处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的图片数量；
S41分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
S421根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；
S422根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。
S431根据所述第k特征点集计算得到第k帧目标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；
S432根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N+1像素坐标信息作为第k+N+1质心信息；
所述步骤S432中所述的第k质心像素坐标的计算公式如下：
Figure PCTCN2017110287-appb-000005
其中，P(k)(j)表示第k帧目标图片信息对应的第k特征点集中第j个特征点的像素坐标，C(k)表示所述第k质心像素坐标；
所述步骤S432中所述的第k+N+1质心像素坐标的计算公式如下；
Figure PCTCN2017110287-appb-000006
其中，P(k+N+1)(j)表示第k+N+1帧图片信息对应的第k+N+1特征点集中第j个特征点的像素坐标，C(k+N+1)表示所述第k+N+1质心像素坐标；
S441计算得到所述第k帧目标图片信息的中心位置像素坐标作为中心位置信息；
S442根据所述第k质心像素坐标、所述第k+N+1质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标，其计算公式如下：
R(k+N+1)＝R(k)-C(k)+C(k+N+1)
其中，R(k+N+1)表示所述第k+N+1帧图片信息的中心位置像素坐标，R(k)表示所述第k帧目标图片信息的中心位置像素坐标，C(k)表示所述第k质心像素坐标，C(k+N+1)所述第k+N+1质心像素坐标。
S50在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
具体的，本实施例具体阐述了如何计算得到所述第k质心像素坐标、所述第k+N+1像素坐标信息以及如何根据所述第k质心像素坐标、所述第k+N+1质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标。
首先，根据上个实施例所述的，在计算得到第k帧目标图片信息所有特征点像素坐标之后，可以根据步骤S433的公式，计算得到第k帧图片信息的质心坐标；比如计算得到第k帧目标图片信息有3个特征点，他们的像素坐标分别是(2,5)、(1,6)、(3,7)，可通过步骤S433的公式，计算得到质心坐标为(2,6)。同理，也可以根据步骤S434中的公式计算得到第k+N+1帧图片信息的质心坐标。
其次，计算得到所述第k帧目标图片信息的中心位置的像素坐标，由于在预测过程中，图片信息的中心坐标位置和该图片中的质心坐标位置总是呈近似的相对距离，这里为了方便，取相等距离，因此可以推出步骤S450中的公式。
根据步骤S450中所述的公式，计算得到第k+N+1帧图片信息的中心位 置像素坐标。根据第k+N+1帧图片信息的中心位置像素坐标以及跟随目标的质心坐标，可以计算得到第k+N+1帧图片中跟随目标的位置具体在哪个地方，从而预测得到第k+N+1帧图片信息。通过上述方法，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置。
最后，在预测之后，机器人可以判断跟随目标具体是如何移动，根据检测的结果调整自身的位置。
如图10，11所示，以通过第一帧frame 1检测结果rect 1(即frame 1中包含人体上半身的部分，如图11中左侧的小矩形框)和采集的第n+1帧预测显示的第2帧为例。
首先，分别计算rect 1中的特征点和第n+1帧整张图的特征点，例如，FAST特征点。
其次，对这两个图片的特征点进行匹配。具体匹配方法为：通过K近邻一致性算法计算两张图的特征点相似度，并基于Ransac随机采样一致算法滤除错误匹配点，最终得到分别属于frame 1的内点集P(1)和frame n+1的内点集P(n+1)，这两组点集是一一对应匹配的。
再次，假设点集个数为n，则通过步骤S432中的公式分别计算P(1)和P(n+1)的质心；之后：预测第2帧图像中，人体上半身的位置；根据相邻帧中，目标关键点中心位置和该目标的矩形框中心位置总是呈近似的相对距离，这里为了方便，取相等距离。因此可以得到步骤S442中的公式，并且rect n+1的矩形框大小等于rect 1。如图11右边所示，预测的结果为虚线矩形框rect n+1，以便于显示模块可以将采集到的第n+1帧图片和预测的虚线矩形框rect n+1作为第2帧展示的图像。
最后，当采集模块采集到下一帧时，重复上述操作，即可不断的展示预测的上半身目标的位置。
预测模块的处理过程中，耗时部分在于计算特征点，而计算特征点只需要依次对图像中各个点及邻域3*3内的像素灰度进行比较，处理速度快(比较算法简单)；检测模块的处理过程中，需要对图片进行大量的卷积操作以提取特征点，其卷积核的参数高达几百万个，大量浮点型运算比较 耗时。所以预测模块的处理速度要远远高于检测模块的检测速度。
如图5所示，本发明提供了一种基于机器人的目标图像显示处理系统的一个实施例，包括：
采集模块1，用于对视频帧图像进行连续采集；
检测模块2，与采集模块1电连接，在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
显示模块4，与所述检测模块2电连接，用于在时间轴第k+N+1帧图片处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的帧图片数量；
预测模块3，与所述检测模块电连接，用于根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；
所述显示模块4，还用于在时间轴第k+N+2至k+2N帧图片处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
具体的，采集模块可以由机器人上安装的摄像头构成，用于采集跟随目标的图像，当跟随目标出现在机器人视野范围内时，机器人开始连续采集含有跟随目标的图片；在采集到第N帧图片信息时，机器人的检测模块刚好检测完第1帧图片信息得到第1帧目标图片信息，并通过显示模块在时间轴上第N+1帧的位置显示检测完的第1帧目标图片。之后，机器人跳过对第N+2帧图片信息到第2N帧图片信息的检测，直接对第N+1帧图片信息进行检测，并在时间轴第2N帧的位置显示检测之后的第N+1帧目标图片信息。
由于机器人在连续采集图片信息，当检测完第1帧图片信息并显示后，根据检测得到的第1帧目标图片信息和采集到的第N+1帧图片信息，预测得到第2帧目标图片信息，并在时间轴第N+2的位置显示预测出来的第2帧目标图片信息。重复执行上述预测功能，可以得到第3帧目标图片信息，直到预测出并显示出第2N-1帧目标预测图片信息通过本实施例提供的预测功能，能够弥补现有技术出现跳频和卡顿的现象。
如图6所示，本发明提供了一种基于机器人的目标图像显示处理系统的另一个实施例，包括：
采集模块1，用于对视频帧图像进行连续采集；
检测模块2，与采集模块1电连接，在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
显示模块4，与所述检测模块2电连接，用于在时间轴第k+N+1帧图片处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的帧图片数量；
预测模块3，与所述检测模块2电连接，用于根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；
所述显示模块4，还用于在时间轴第k+N+2至k+2N帧图片处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置；
预测模块3包括：
计算子模块31，用于分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
匹配子模块32，与所述计算子模块31电连接，对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；
所述计算子模块31，还用于根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；
预测子模块33，用于根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+1帧目标预测图片信息。
在本实施例中，对如何根据检测得到的第k帧目标图片信息，和采集到的第k+N+1帧图片信息，预测得到第k+N+1帧图片跟随目标的位置进行 了具体的阐述。首先，图像处理模块运用现有的一些对图片中特征点的处理方法，分别得到所述第k帧目标图片信息对应的第k特征点信息和采集到的第k+N+1帧图片信息对应的第k+N+1特征点信息，如每个特征点在图片中的像素坐标；其次对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集，这两个特征点集中的特征点都是两两对应的关系；再次，根据所述第k特征点集、所述第k+N+1特征点集，计算模块分别计算出所述第k特征点集的第k质心信息、第k+N+1特征点集的第k+N+1质心信息，所述的特征信息包括质心在图片中的像素坐标；最后，根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+N+1帧目标预测图片信息。
如图7所示，本发明提供了一种基于机器人的目标图像显示处理系统的另一个实施例，包括：
采集模块1，用于对视频帧图像进行连续采集；
检测模块2，与采集模块1电连接，在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
显示模块4，与所述检测模块2电连接，用于在时间轴第k+N+1帧图片处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置；N为一个检测周期内所采集的帧图片数量；
预测模块3，与所述检测模块2电连接，用于根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；
所述显示模块4，还用于在时间轴第k+N+2至k+2N帧图片处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置；
预测模块3包括：
计算子模块31，用于分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
匹配子模块32，与所述计算子模块31电连接，对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；
所述计算子模块31，还用于根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；
预测子模块33，用于根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+1帧目标预测图片信息；
匹配子模块32包括：
图片处理单元321，用于根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；
所述图片处理单元321，还用于根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集；
所述计算子模块31，还用于根据所述第k特征点集计算得到第k帧目标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；
所述计算子模块31，还用于根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N像素坐标信息作为第k+N+1质心信息。
所述计算子模块，还用于计算所述第k质心像素坐标，计算公式如下：
Figure PCTCN2017110287-appb-000007
其中，P(k)(j)表示第k帧目标图片对应的第k特征点集中第j个特征点的像素坐标，C(k)表示所述第k质心像素坐标；
所述计算子模块，还用于计算所述第k+N+1质心像素坐标，计算公式如下；
Figure PCTCN2017110287-appb-000008
其中，P(k+N+1)(j)表示第k+N+1帧图片信息对应的第k+N+1特征点集中第j个特征点的像素坐标，C(k+N+1)表示所述第k+N+1质心像素坐标。
所述计算子模块31，还用于计算得到所述第k帧目标图片信息的中心位置像素坐标作为中心位置信息；
所述计算子模块31，还用于根据所述第k质心像素坐标、所述第k+N+1质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标，其计算公式如下：
R(k+N+1)＝R(k)-C(k)+C(k+N+1)
其中，R(k+N+1)表示所述第k+N+1帧图片信息的中心位置像素坐标，R(k)表示所述第k帧目标图片信息的中心位置像素坐标，C(k)表示所述第k质心像素坐标，C(k+N+1)所述第k+N+1质心像素坐标。
具体的，本实施例具体阐述了如何计算得到所述第k质心像素坐标、所述第k+N+1像素坐标信息以及如何根据所述第k质心像素坐标、所述第k+N+1质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标。
具体的，本实施例主要阐述了如何对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集的过程；以及如何根据所述第k特征点集、所述第k+N+1特征点集，分别计算出所述第k特征点集的第k质心信息、第k+N+1特征点集的第k+N+1质心信息的过程。
首先，匹配子模块可以通过特征点检测方法分别得到所述第k帧目标图片信息对应的第k特征点信息和采集到的第k+N+1帧图片信息对应的第k+N+1特征点信息，如用FAST特征点检测方法得到第k帧目标图片信息对应的第k特征点信息和采集到的第k+N+1帧图片信息对应的第k+N+1特征点信息；
其次，根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1 特征点信息的相似度，图片处理模块可以得到对应的匹配点。在实际情况中，两张图片不一样，计算得到的特征点也不全是相同的，因此，需要根据K邻近一致性算法计算第k特征点信息和所述第k+N+1特征点信息的相似度，得到两张图片对应的匹配点。
再次，根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。在实际情况中，图片计算出来的特征点除了跟随目标的特征点以外，还有其他背景图片中的一些特征点，这些特征点其实并不是检测和预测所需要的特征点，所以需要根据Ransac随机采样一致算法滤除错误匹配点，得到正确的特征点，也就是所述的第k特征点集和所述的第k+N+1特征点集。
之后，根据所述第k特征点集、所述第k+N+1特征点集分别计算得到第k帧目标图片信息内所有特征点的像素坐标、第k+N+1帧图片信息内所有特征点的像素坐标；所述的像素坐标可以以图片的一个角为顶点建立坐标系，然后通过计算每个特征点在坐标系中的位置得到每个特征点对应的像素坐标。
最后，根据所述第k帧目标图片信息内所有特征点的像素坐标、所述第k+N+1帧图片信息内所有特征点的像素坐标，计算得到所述第k质心像素坐标、所述第k+N+1像素坐标信息；
本实施例中，在计算得到第k帧目标图片信息所有特征点像素坐标之后，可以根据步骤S433的公式，计算得到第k帧图片信息的质心坐标；比如计算得到第k帧目标图片信息有3个特征点，他们的像素坐标分别是(2,5)、(1,6)、(3,7)，可通过公式，计算得到质心坐标为(2,6)。同理，也可以根据步骤S434中的公式计算得到第k+N帧图片信息的质心坐标。
之后，可以计算得到所述第k帧目标图片信息的中心位置的像素坐标，由于在预测过程中，图片信息的中心坐标位置和该图片中的质心坐标位置总是呈近似的相对距离，这里为了方便，取相等距离，因此可以得到公式：R(k)-C(k)＝R(k+N+1)-C(k+N+1)，从而可以推到出公式：R(k+N+1)＝R(k)-C(k)-C(k+N+1)。
根据步骤S450中所述的公式，计算得到第k+N+1帧图片信息的中心位置像素坐标。根据第k+N+1帧图片信息的中心位置像素坐标，就可以计算得到第k+N+1帧图片信息的位置具体在哪个位置，从而预测得到第k+N+1帧图片中跟随目标的位置。通过上述方法，还可以预测得到第k+N+2帧目标预测图片信息～第k+2N-1帧目标预测图片信息。
应当说明的是，上述实施例均可根据需要自由组合。以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。

Claims (16)

  1. 一种基于机器人的目标图像显示处理方法，其特征在于，包括步骤：
    S10对视频帧图像进行连续采集；
    S20在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
    S30在时间轴第k+N+1帧处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置，N为一个检测周期内所采集的图片数量；
    S40根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；
    S50在时间轴第k+N+2至k+2N帧处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
  2. 如权利要求1所述的一种基于机器人的目标图像显示处理方法，其特征在于，步骤S40中根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置具体为：
    根据所述跟随目标在所述第k帧图片的位置信息，预测在第k+N+1帧图片中跟随目标的位置，以及根据第k+N+i帧图片中跟随目标的位置，预测k+N+i+1帧图片中跟随目标的位置；其中i分别为1～N-2的正整数。
  3. 如权利要求1所述的一种基于机器人的目标图像显示处理方法，其特征在于，该方法进一步包括：
    在步骤S30中显示所述第k帧图片的同时，对采集到的第k+N帧图片进行检测，以及在步骤S50之后，在时间轴第k+2N+1帧图片处显示所述k+N帧图片以及所述检测到的跟随目标的位置。
  4. 如权利要求1或2所述的一种基于机器人的目标图像显示处理方法，其特征在于，根据所述跟随目标在所述第k帧图片的位置信息，预测在第k+N+1帧中跟随目标的位置具体为：
    S41分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
    S42对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；
    S43根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；
    S44根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+N+1帧图片中所述跟随目标的位置。
  5. 如权利要求4所述的一种基于机器人的目标图像显示处理方法，其特征在于，S42中对所述第k特征点信息和所述第k+N+1特征点信息进行匹配的步骤具体包括：
    S421根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；
    S422根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。
  6. 如权利要求4所述的一种基于机器人的目标图像显示处理方法，其特征在于，步骤S43中计算所述第k+N+1特征点集的第k+N+1质心信息具体包括：
    S431根据所述第k特征点集计算得到第k帧目标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；
    S432根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N+1像素坐标信息作为第k+N+1质心信息。
  7. 如权利要求6所述的一种基于机器人的目标图像显示处理方法，其特征在于：
    所述步骤S432中所述的第k质心像素坐标的计算公式如下：
    Figure PCTCN2017110287-appb-100001
    其中，P(k)(j)表示第k帧目标图片信息对应的第k特征点集中第j个特征点的像素坐标，C(k)表示所述第k质心像素坐标；
    所述步骤S432中所述的第k+N+1质心像素坐标的计算公式如下；
    Figure PCTCN2017110287-appb-100002
    其中，P(k+N+1)(j)表示第k+N+1帧图片信息对应的第k+N+1特征点集中第j个特征点的像素坐标，C(k+N+1)表示所述第k+N+1质心像素坐标。
  8. 如权利要求7所述的一种基于机器人的目标图像显示处理方法，其特征在于，步骤S44中预测得到第k+N+1帧图片中跟随目标的位置具体为：
    S441计算得到所述第k帧目标图片信息的中心位置像素坐标作为中心位置信息；
    S442根据所述第k质心像素坐标、所述第k+N+1质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标，其计算公式如下：
    R(k+N+1)＝R(k)-C(k)+C(k+N+1)
    其中，R(k+N+1)表示所述第k+N+1帧图片信息的中心位置像素坐标，R(k)表示所述第k帧目标图片信息的中心位置像素坐标，C(k)表示所述第k质心像素坐标，C(k+N+1)所述第k+N+1质心像素坐标。
  9. 一种基于机器人的目标图像显示处理系统，其特征在于，包括：
    采集模块，用于对视频帧图像进行连续采集；
    检测模块，与采集模块电连接，在采集到的第k帧图片中检测到跟随目标时，检测所述跟随目标在所述第k帧图片的位置信息；
    显示模块，与所述检测模块电连接，用于在时间轴第k+N+1帧图片处显示所述第k帧图片，并标记出所述跟随目标在所述第k帧图片的位置，N为一个检测周期内所采集的帧图片数量；
    预测模块，与所述检测模块电连接，用于根据所述跟随目标在所述第k帧图片的位置信息，依次预测在采集到的第k+N+1至第k+2N-1帧图片中跟随目标的位置；
    所述显示模块，还用于在时间轴第k+N+2至k+2N帧图片处依次显示所述k+N+1至k+2N-1帧图片以及所述预测的跟随目标的位置。
  10. 如权利要求9所述的一种基于机器人的目标图像显示处理系统，其特征在于：
    所述预测模块还用于根据所述跟随目标在所述第k帧图片的位置信息，预测在第k+N+1帧中跟随目标的位置，以及根据第k+N+i帧中跟随目标的位置，预测k+N+i+1帧中跟随目标的位置；其中i分别为1～N-2的正整数。
  11. 如权利要求9所述的一种基于机器人的目标图像显示处理系统，其特征在于：
    所述检测模块还用于显示所述第k帧图片的同时，对采集到的第k+N帧图片进行检测；
    所述显示模块还用于在时间轴第k+2N+1帧图片处显示所述k+N帧图片以及所述检测到的跟随目标的位置。
  12. 如权利要求9或10所述的一种基于机器人的目标图像显示处理系统，其特征在于，所述预测模块还包括：
    计算子模块，用于分别计算出所述第k帧图片跟随目标对应的第k特征点信息和采集到的第k+N+1帧图片跟随目标对应的第k+N+1特征点信息；
    匹配子模块，与所述计算子模块电连接，对所述第k特征点信息和所述第k+N+1特征点信息进行匹配，分别得到第k特征点集、第k+N+1特征点集；
    所述计算子模块，还用于根据所述第k特征点集计算出所述第k特征点集的第k质心信息，以及根据所述第k+N+1特征点集计算出所述第k+N+1特征点集的第k+N+1质心信息；
    预测子模块，用于根据所述第k质心信息、所述第k+N+1质心信息以及所述第k帧目标图片信息的中心位置信息，预测得到第k+1帧目标预测图片信息。
  13. 如权利要求12所述的一种基于机器人的目标图像显示处理系统，其特征在于，所述匹配子模块还包括：
    图片处理单元，用于根据K邻近一致性算法计算所述第k特征点信息和所述第k+N+1特征点信息的相似度，得到对应的匹配点；
    所述图片处理单元，还用于根据Ransac随机采样一致算法滤除错误匹配点，得到所述第k特征点集和所述第k+N+1特征点集。
  14. 如权利要求12所述的一种基于机器人的目标图像显示处理系统，其特征在于：
    所述计算子模块，还用于根据所述第k特征点集计算得到第k帧目 标图片信息内所有特征点的像素坐标，以及根据所述第k+N+1特征点集计算得到第k+N+1帧图片信息内所有特征点的像素坐标；
    所述计算子模块，还用于根据所述第k帧目标图片信息内所有特征点的像素坐标计算得到所述第k质心像素坐标作为第k质心信息，以及根据所述第k+N+1帧图片信息内所有特征点的像素坐标计算得到所述第k+N像素坐标信息作为第k+N+1质心信息。
  15. 如权利要求14所述的一种基于机器人的目标图像显示处理系统，其特征在于：
    所述计算子模块，还用于计算所述第k质心像素坐标，计算公式如下：
    Figure PCTCN2017110287-appb-100003
    其中，P(k)(j)表示第k帧目标图片对应的第k特征点集中第j个特征点的像素坐标，C(k)表示所述第k质心像素坐标；
    所述计算子模块，还用于计算所述第k+N+1质心像素坐标，计算公式如下；
    Figure PCTCN2017110287-appb-100004
    其中，P(k+N+1)(j)表示第k+N+1帧图片信息对应的第k+N+1特征点集中第j个特征点的像素坐标，C(k+N+1)表示所述第k+N+1质心像素坐标。
  16. 如权利要求15所述的一种基于机器人的目标图像显示处理系统，其特征在于：
    所述计算子模块，还用于计算得到所述第k帧目标图片信息的中心位置像素坐标作为中心位置信息；
    所述计算子模块，还用于根据所述第k质心像素坐标、所述第k+N+1 质心像素坐标、所述第k帧图片信息的中心位置像素坐标，计算得到所述第k+N+1帧图片信息的中心位置像素坐标，其计算公式如下：
    R(k+N+1)＝R(k)-C(k)+C(k+N+1)
    其中，R(k+N+1)表示所述第k+N+1帧图片信息的中心位置像素坐标，R(k)表示所述第k帧目标图片信息的中心位置像素坐标，C(k)表示所述第k质心像素坐标，C(k+N+1)所述第k+N+1质心像素坐标。
PCT/CN2017/110287 2017-09-29 2017-11-09 一种基于机器人的目标图像显示处理方法及系统 WO2019061713A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/972,073 US10733739B2 (en) 2017-09-29 2018-05-04 Method and system for displaying target image based on robot

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710911965.XA CN107742303B (zh) 2017-09-29 2017-09-29 一种基于机器人的目标图像显示处理方法及系统
CN201710911965.X 2017-09-29

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/972,073 Continuation US10733739B2 (en) 2017-09-29 2018-05-04 Method and system for displaying target image based on robot

Publications (1)

Publication Number Publication Date
WO2019061713A1 true WO2019061713A1 (zh) 2019-04-04

Family

ID=61236440

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/110287 WO2019061713A1 (zh) 2017-09-29 2017-11-09 一种基于机器人的目标图像显示处理方法及系统

Country Status (2)

Country Link
CN (1) CN107742303B (zh)
WO (1) WO2019061713A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109141395B (zh) * 2018-07-10 2020-06-09 深圳市无限动力发展有限公司 一种基于视觉回环校准陀螺仪的扫地机定位方法及装置
CN110503042B (zh) * 2019-08-23 2022-04-19 Oppo广东移动通信有限公司 图像处理方法、装置以及电子设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102449427A (zh) * 2010-02-19 2012-05-09 松下电器产业株式会社 物体位置修正装置、物体位置修正方法及物体位置修正程序
CN103578116A (zh) * 2012-07-23 2014-02-12 三星泰科威株式会社 用于跟踪对象的设备和方法
CN103810460A (zh) * 2012-11-09 2014-05-21 株式会社理光 对象跟踪方法和装置
CN105938622A (zh) * 2015-03-02 2016-09-14 佳能株式会社 检测运动图像中的物体的方法和装置
US20170228887A1 (en) * 2016-02-10 2017-08-10 Canon Kabushiki Kaisha Image capturing apparatus, tracking device, control method, and storage medium

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101420606A (zh) * 2007-10-23 2009-04-29 青岛海信电器股份有限公司 图像处理方法和装置
WO2013101892A1 (en) * 2011-12-29 2013-07-04 Quest Diagnostics Investments Incorporated Methods for detecting hyperglycosylated human chorionic gonadotropin (hcg-h)
CN105913028B (zh) * 2016-04-13 2020-12-25 华南师范大学 一种基于face++平台的人脸跟踪方法及其装置
CN106125087B (zh) * 2016-06-15 2018-10-30 清研华宇智能机器人（天津）有限责任公司 基于激光雷达的舞蹈机器人室内行人跟踪方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102449427A (zh) * 2010-02-19 2012-05-09 松下电器产业株式会社 物体位置修正装置、物体位置修正方法及物体位置修正程序
CN103578116A (zh) * 2012-07-23 2014-02-12 三星泰科威株式会社 用于跟踪对象的设备和方法
CN103810460A (zh) * 2012-11-09 2014-05-21 株式会社理光 对象跟踪方法和装置
CN105938622A (zh) * 2015-03-02 2016-09-14 佳能株式会社 检测运动图像中的物体的方法和装置
US20170228887A1 (en) * 2016-02-10 2017-08-10 Canon Kabushiki Kaisha Image capturing apparatus, tracking device, control method, and storage medium

Also Published As

Publication number Publication date
CN107742303B (zh) 2021-05-25
CN107742303A (zh) 2018-02-27

Similar Documents

Publication Publication Date Title
US10733739B2 (en) Method and system for displaying target image based on robot
US10445887B2 (en) Tracking processing device and tracking processing system provided with same, and tracking processing method
CN111612820B (zh) 多目标跟踪方法、特征提取模型的训练方法和装置
CN101593022B (zh) 一种基于指端跟踪的快速人机交互方法
CN110598559B (zh) 检测运动方向的方法、装置、计算机设备和存储介质
JP2014002744A (ja) イベントベースイメージ処理装置及びその装置を用いた方法
TWI640931B (zh) 影像目標追蹤方法及裝置
JP2016106668A (ja) 情報処理装置、情報処理方法、およびプログラム
CN103745485A (zh) 判断物体静止或运动的方法及系统
CN104821010A (zh) 基于双目视觉的人手三维信息实时提取方法及系统
JP2010182014A (ja) ジェスチャ認識装置、その方法及びそのプログラム
CN103679130B (zh) 手追踪方法、手追踪设备和手势识别系统
JP2012212373A (ja) 画像処理装置、画像処理方法及びプログラム
WO2019061713A1 (zh) 一种基于机器人的目标图像显示处理方法及系统
WO2020143499A1 (zh) 一种基于动态视觉传感器的角点检测方法
CN104376323A (zh) 一种确定目标距离的方法及装置
CN112637587A (zh) 坏点检测方法及装置
JP6058720B2 (ja) 情報出力装置、検知装置、プログラム及び情報出力方法
WO2024000558A1 (zh) 目标跟踪方法、目标跟踪系统和电子设备
CN102131078B (zh) 一种视频图像校正方法和系统
CN107341818A (zh) 用于触摸屏响应性能测试的图像分析算法
CN108734065A (zh) 一种手势图像获取设备及方法
Wu et al. Moving target detection based on improved three frame difference and visual background extractor
JP2020043544A (ja) 撮像装置およびその制御方法、プログラムならびに記憶媒体
CN108073271A (zh) 基于预定区域识别手部区域的方法及装置

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: 17926469

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: 17926469

Country of ref document: EP

Kind code of ref document: A1