US20130058535A1 - Detection of objects in an image using self similarities - Google Patents

Detection of objects in an image using self similarities Download PDF

Info

Publication number
US20130058535A1
US20130058535A1 US13/697,212 US201113697212A US2013058535A1 US 20130058535 A1 US20130058535 A1 US 20130058535A1 US 201113697212 A US201113697212 A US 201113697212A US 2013058535 A1 US2013058535 A1 US 2013058535A1
Authority
US
United States
Prior art keywords
self
similarity
image
pixels
detection window
Prior art date
Legal status (The legal status 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 status listed.)
Abandoned
Application number
US13/697,212
Other languages
English (en)
Inventor
Gabriel Othmezouri
Ichiro Sakata
Bernt Schiele
Stefan Walk
Nikodem Majer
Konrad Schindler
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Toyota Motor Europe NV SA
Technische Universitaet Darmstadt
Original Assignee
Toyota Motor Europe NV SA
Technische Universitaet Darmstadt
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 Toyota Motor Europe NV SA, Technische Universitaet Darmstadt filed Critical Toyota Motor Europe NV SA
Publication of US20130058535A1 publication Critical patent/US20130058535A1/en
Abandoned legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G06V10/507—Summing image-intensity values; Histogram projection analysis
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows

Definitions

  • This invention relates to apparatus and methods for image processing to detect objects such as humans, and to corresponding computer programs for carrying out such methods and to memory devices storing the computer programs and also to corresponding integrated circuits.
  • Pedestrian detection has been a focus of recent research due to its importance for practical applications such as automotive safety [see refs 11, 8] and visual surveillance [23].
  • the most successful model to date for “normal” pedestrians, who are usually standing or walking upright, is still a monolithic global descriptor for the entire search window.
  • feature extraction classification
  • non-maxima suppression there are three main steps which can be varied to gain performance: feature extraction, classification, and non-maxima suppression.
  • the most common features extracted from the raw image data are variants of the HOG framework, i.e. local histograms of gradients and (relative) optic flow [3, 4, 10, 24, 27], and different flavors of generalized Haar wavelets, e.g. [6, 23].
  • Competitive classifiers we know of employ statistical learning techniques to learn the mapping from features to scores (indicating the likelihood of a pedestrian being present)—usually either support vector machines [3, 13, 17, 19, 27] or some variant of boosting [23, 27, 28, 30].
  • An object of the invention is to provide apparatus and methods for image processing to detect objects such as humans, and to corresponding computer programs for carrying out such methods and to corresponding integrated circuits. According to a first aspect, the invention provides:
  • An image processor for detection of an object in an image or sequence of images, each image being formed of pixels, and the image processor comprising: a window selector for choosing a detection window within the image, a self similarity computation part for determining self-similarity information for a group of the pixels in any part of the detection window, to represent an amount of self-similarity of that group to other groups in any other part of the detector window, and for repeating the determination for groups in all parts of the detection window, to generate a global self similarity descriptor for the chosen detection window, and a classifier for classifying whether the object is present in the detection window of the image from the global self-similarity descriptor for that detection window.
  • Embodiments of the invention can have any other features added, some such additional features are set out in dependent claims and described in more detail below.
  • FIG. 1 shows an image processor according to a first embodiment
  • FIG. 2 shows method steps according to an embodiment
  • FIG. 3 shows another embodiment
  • FIG. 4 shows views of a window of an image and self similarity information from four cells in different parts of the image
  • FIG. 5 shows an image processor according to an embodiment having a motion detector
  • FIG. 6 shows steps in determining gradient information according to an embodiment
  • FIG. 7 shows steps in determining self similarity information according to an embodiment
  • FIG. 8 shows an example of an image showing detection windows
  • FIG. 9 shows steps according to another embodiment.
  • references to a signal can encompass any kind of signal in any medium, and so can encompass an electrical or optical or wireless signal or other signal for example.
  • References to analyzing can encompass processing a signal in any way to derive or enhance information about the material.
  • References to a processor can encompass any means for processing signals or data in any form and so can encompass for example a personal computer, a microprocessor, analog circuitry, application specific integrated circuits, software for the same, and so on.
  • At least some embodiments of the present invention have a new feature based on self-similarity of low level features, in some particular embodiments color histograms from different sub-regions within the detector window.
  • This feature termed CSS, captures pairwise statistics of spatially localized color distributions, thus being independent of the actual color of a specific example. It is global similarities which are found, in the sense of similarities of a given sub-region with any other part of the detection window, not just the parts adjacent to the sub-region.
  • the self-similarity enables representation of properties like “the color distributions on the left and right shoulder usually exhibit high similarity”, independently of the actual color distribution, which may vary from person to person depending on their clothing.
  • Adding CSS can improve state-of-the-art classification performance for both static images and image sequences.
  • the new feature is particularly powerful for static images, and hence also valuable for applications such as content-based image retrieval. It also yields a consistent improvement on image sequences, in combination with motion detection in the form of optic flow.
  • Instance-specific color information was recently used in the form of implicit local segmentation features [15], encoding gradients of distances w.r.t. two local color distribution models (“foreground” and “background”). Only few authors have advocated the use of self-similarity as a feature.
  • [20] encodes the local self-similarity of raw image patches in a log-polar binned descriptor. They demonstrate superior performance over gradient features in a template matching task, which can include matching particular human poses.
  • the authors propose self-similarity descriptors over feature time series for human action recognition, observing good viewpoint invariance of the descriptor.
  • [21] proposed a representation where color similarity is computed at the pixel level, assuming a Gaussian conditional color distribution.
  • HOG Histograms of oriented gradients are a known feature for object detection, first proposed in [3]. They collect gradient information in local cells into histograms using trilinear interpolation, and normalize overlapping blocks composed of neighbouring cells. Interpolation, local normalization and histogram binning make the representation robust to changes in lighting conditions and small variations in pose. HOG can optionally be enhanced by Local Binary Patterns (LBP) [24].
  • LBP Local Binary Patterns
  • FIG. 1 shows an image processor according to an embodiment.
  • FIG. 2 shows steps carried out by this or other embodiments.
  • the image processor can be implemented as for example one or more integrated circuits having hardware such as circuit blocks dedicated to each of the parts shown, or can be implemented for example as software modules executed by a general purpose processor in sequence, as in a server.
  • the parts shown include a selector 20 for receiving an input image or image stream (such as frames of a video, in real time or non real time) from an image source device 5 , and selecting a detection window, and within that window, selecting groups of pixels to be processed.
  • the groups can be e.g. 6 ⁇ 6 or 8 ⁇ 8 pixels or different sizes. They need not be square, and can be rectangular or other regular or irregular shape.
  • the self similarity computation part determines self similarity information for a group of the pixels in any part of the detection window, to represent an amount of self-similarity of that group to other groups in any other part of the detector window, and repeats the determination for groups in all parts of the detection window, to generate a global self similarity descriptor for the chosen detection window. Again this can be implemented in various ways, and an example will be described below in more detail with reference to FIG. 7 .
  • the self similarity information for different parts of the window can be determined in parallel or sequentially and are fed to a classifier 50 .
  • This can be a binary decision or can produce a score, and can be carried out in various ways, and an example is described in more detail below.
  • a device 55 can be provided for interpreting or taking action based on the classifier score or decision. This can be for example a vehicle control system, or driver assistance system, a robotic system, a surveillance system for detecting intruders and so on.
  • FIG. 2 shows steps in operating the image processor of FIG. 1 or of other embodiments.
  • a window is selected, and a first group of pixels is selected.
  • self similarity information is determined for a group of the pixels in any part of the detection window, to represent an amount of self-similarity of that group to other groups in any other part of the detector window.
  • this is repeated for a next group, if the global self-similarities have not been processed for all parts of the window. Once it has been done for all parts of the detection window, to generate a global self similarity descriptor for the chosen detection window, this descriptor can be used by the classifier.
  • the classification of whether an object is present in the image is made from the descriptor.
  • Additional features can include the self similarity information comprising an amount of self-similarity of colours of pixels of the group. This is one of several useful features which can help distinguish objects such as humans in particular.
  • the image processor can have a part ( 42 ) arranged to determine a distribution of colours of the pixels of the groups, and the self similarity information comprising an amount of self-similarity of the colour distributions. This is another feature which can help distinguish objects such as humans in particular.
  • a part ( 30 ) for determining gradient information by determining a distribution of intensity gradients in a cell of pixels, and for inputting such gradient information for cells over all parts of the detection window to the classifier, the classifier additionally being arranged to use the gradient information to classify whether the object is present.
  • the gradient information can be complementary to the self similarity information in many cases, and hence provide more distinctive information to the classifier to help enable better classification.
  • the image processor can have a part arranged to determine a flow of the groups in terms of motion vectors of the pixels of the groups over successive images in a sequence of images, and the self-similarity information comprising an amount of self-similarity of the flow. This is another feature which can help distinguish moving objects such as pedestrians.
  • the self-similarity computation part can have a histogram generator ( 44 ) arranged to determine a histogram of values for a feature of pixels in the group, by using interpolation.
  • a histogram generator 44
  • Such interpolation enables some data compression, to reduce computation load for subsequent steps, and enable faster or cheaper processing.
  • the self similarity computation part can have a part ( 46 ) arranged to determine similarities between histograms for different groups of pixels in the detection window by a histogram intersection. Histogram intersection is one of a number of ways of determining similarities and proves to be particularly efficient and effective.
  • the image processor can comprise a motion detection part ( 70 ) for detecting motion vectors for parts of the image, and the classifier part being arranged to classify based also on the motion vectors of parts in the detection window.
  • a motion detection part 70
  • the classifier part being arranged to classify based also on the motion vectors of parts in the detection window.
  • Such motion information is also useful to distinguish humans in some situations and is often complementary to the self similarity information.
  • the image processor can have a combiner part ( 60 ) for combining the similarity information and the distributions of intensity gradients before input to the classifier.
  • the classifier could operate on the information separately, it is usually more efficient to combine the information first.
  • FIG. 3 shows an embodiment similar to that of FIG. 1 , but showing more details of one way to implement part 40 , for computing the global self-similarity information, showing some of the functions it can carry out.
  • the feature for the self similarity information is colour distribution, so there is shown a step of determining colour distribution for a group of pixels 42 .
  • a histogram is generated, optionally using interpolation 44 to reduce the amount of data and reduce aliasing effects.
  • the histograms for different groups are typically stored and retrieved as needed by a step 46 of determining the similarity between the histogram of a given group and other groups anywhere in the detection window.
  • the result can be a large number of similarity values, which can be regarded as a multidimensional matrix or vector, having as many dimensions as there are pairs of groups (that is G*(G ⁇ 1)/2) where G is the number of groups.
  • the groups can be adjacent or overlapping, or spread apart, but should be chosen from different parts of the window so that the self-similarities are global within the window, and not local in the sense of being only relative to other groups adjacent to the group being considered. This can lead to a normalizing step 48 for normalizing the output vector to account for conditions such as camera noise/image artifacts/different amounts of clutter, or any other causes of some images having overall a lower “self-similarity” for example.
  • a combiner 60 can be provided for combining the self-similarity information with other information such as gradient information. This implies there is a compatible data format for both, so the gradient information can be in the form of a vector which can be stacked with the vector of self similarity information, for input to the classifier.
  • the combined data can be fed to the classifier 50 for the decision or scoring of whether the object has been detected. This can be repeated for other detection windows within the image.
  • the other windows can be chosen for example by sliding, by zooming to alter the scale, or by seeking areas of interest, using known algorithms which need not be described here.
  • FIG. 4 Views of Self Similarity Information
  • FIG. 4 shows an example detection window within an image on the left hand side and shows on its right, four different views of self similarity information determined for four particular points in this example.
  • the self similarity information is computed at marked cell positions using HSV+histogram intersection methods as discussed in more detail below. Cells with higher similarity are brighter.
  • the first on the left represents a colour similarity of every other group of pixels with a group located at the head of the human in the image.
  • the second view is the colour similarity of every other group of pixels with a group located at the stomach of the human in the image.
  • the third view is the colour similarity of every other group of pixels with a group located at an upper leg region of the human in the image.
  • the fourth view is the colour similarity of every other group of pixels with a group located off the human, at a point showing water in the background in the image. Note how self-similarity highlights and distinguishes relevant parts like clothing and visible skin regions.
  • FIG. 5 shows an embodiment similar to that of FIG. 1 .
  • a window/cell/group selector part 21 selects a detection window within an image, and within that window, selects groups or cells of pixels.
  • the window selection can be by sliding, scaling, or finding an area of interest for example.
  • the cells and groups can be the same size, e.g. 6 ⁇ 6 or 8 ⁇ 8 pixels or different sizes. They need not be square, and can be rectangular or other regular or irregular shape.
  • Cells are fed to the gradient computation part 30 .
  • Groups are sent to the self similarity computation part 40 . Groups and cells are so named to show that different pixels can be sent to different parts for processing, though in principle the same cells could be sent to both parts shown.
  • the gradient computation part determines gradient information such as a steepness of the gradient of intensity, and an orientation for that cell. This can be intensity of brightness or intensity of colours for example. Various algorithms can be used to implement this part, an example is described in more detail below.
  • the global self similarity computation part determines self similarity information for the group, relative to any other parts of the window. Again this can be implemented in various ways, and an example will be described below in more detail with reference to FIG. 7 .
  • the gradient information and the self similarity information can be determined in parallel or sequentially and are both fed to a classifier 50 .
  • a motion detector part 70 can be added, which can determine motion information such as optic flow for a given cell or group, based on frames (preferably consecutive frames). This can be implemented in various ways following established practice, and so will not be described in more detail here. A possible enhancement to this part is described below.
  • FIG. 6 Steps in Determining Gradient Information
  • steps are shown for one way to implement the gradient computation part 30 for determining gradient information in the form of distribution of intensity gradients in each cell.
  • a window is divided into cells of 8 ⁇ 8 pixels.
  • a gradient value for that cell is determined and the orientation of the gradient is determined, from the 64 pixel values, of intensity or colour values as appropriate.
  • these values are separated into 9 bins, one for each of the different orientations.
  • these steps are repeated for other cells, and the bins for different cells are grouped into 2 ⁇ 2 blocks of cells, overlapping by one cell for example. Normalization at step 250 is carried out on a block basis. Other ways of implementation can be envisaged.
  • FIG. 7 Steps in Determining Global Self Similarity Information
  • steps are shown for one way to implement the step 110 of determining global self similarity information.
  • the window is divided into 8 ⁇ 8 groups of pixels.
  • trilinear interpolation is used to compute 128 local colour histograms from 128 8 ⁇ 8 groups of pixels, and/or flow histograms as appropriate. Colour is represented in the well known HSV format, though other colour representations could be used. Flow can be represented as motion vectors.
  • pairs of these histograms are compared to determine a value for similarity. The histogram intersection method is used though other methods could be used. If all 128 are compared to all others, this results in a 8128 dimensional vector of similarity values.
  • L2 normalization is applied to this vector. Results in the form of normalized vectors are output to the combiner or classifier.
  • FIG. 8 Example of Image Showing Detection Windows
  • This figure shows an example of an image containing many objects, some of which are humans, and some are overlapping with others giving rise to occlusions. Detection windows around each possible human are shown. These may be selected based on sliding a window over the image and comparing scores of different windows, and windows at different scales.
  • a window is moved over an image, by sliding, scaling or seeking an area of interest.
  • a global self-similarity descriptor and optionally other features are obtained for that window.
  • a classifier is used to generate a score or a decision for each window.
  • scores for different windows are compared, and this may be used to decide where to move and/or scale the window. This may lead to repeating steps 400 to 430 .
  • step 440 may be carried out, using the scores for different windows to locate an object and take action based on the scores and the location, such as to control a vehicle or robot, or raise an alarm in a surveillance system for example.
  • the 4 squares display the encoding for one histogram each.
  • the optical flow corresponding to the pixel at the ith row and jth column of the upper left cell is subtracted from the one at the corresponding position of the lower left cell, and the resulting vector votes into a histogram as in the original HOF scheme.
  • IMHd2 provides a dimensionality reduction of 44% (2520 instead of 4536 values per window), without changing performance significantly.
  • Embodiments of the present invention can combine two of these ideas and use second order statistics of colors for example as an additional feature.
  • Color by itself is of limited use, because colors vary across the entire spectrum both for people (respectively their clothing) and for the background, and because of the essentially unsolved color constancy problem.
  • people do exhibit some structure, in that colors are locally similar—for example (see FIG. 4 ) the skin color of a specific person is similar on their two arms and face, and the same is true for most people's clothing. Therefore, we encode color self-similarities within the descriptor window, i.e. similarities between colors in different sub-regions.
  • self-similarity is not limited to color histograms and directly generalizes to arbitrary localized subfeatures within the detector window.
  • features such as gradient orientation in the form of HOG blocks or motion detection features such as flow histograms.
  • Linear SVMs remain a popular choice for people detection because of their good performance and speed.
  • Nonlinear kernels typically bring some improvement, but commonly the time required to classify an example is linear in the number of support vectors, which is intractable in practice.
  • An exception is the (histogram) intersection kernel (HIK) [14], which can be computed exactly in logarithmic time, or approximately in constant time, while consistently outperforming the linear kernel.
  • AdaBoost Viola et al. [23] used AdaBoost in their work on pedestrian detection. However, it has since been shown that AdaBoost does not perform well on challenging datasets with multiple viewpoints [27]. MPLBoost remedies some of the problems by learning multiple (strong) classifiers in parallel. The final score is then the maximum score over all classifiers, allowing individual classifiers to focus on specific regions of the feature space without degrading the overall classification performance.
  • Performance was measured on the “reasonable” subset of Caltech Pedestrians, which is the most popular portion of the data. It consists of pedestrians of ⁇ 50 pixels in height, who are fully visible or less than 35% occluded.
  • Our detector in its strongest incarnation, using HOG, HOF and CSS in a HIKSVM (HOGF+CSS), outperforms the previous top performers—the channel features (ChnFtrs) of [5] and the latent SVM (LatSvm ⁇ V2) of [10]—by a large margin:
  • HOG with HIKSVM is on par with the state of the art [5, 10], which illustrates the effect of correct bootstrapping, and the importance of careful implementation. We did not tune our detector to the dataset. Still, to make sure the performance gain is not dataset-specific, we have verified that our detector outperforms the original HOG implementation [3] also on INRIAPerson (also note that adding CSS provides an improvement for HOG+LBP). HOG+CSS is consistently better than HOG alone, providing an improvement of 5.9% at 0.1 fppi, which indicates that color self-similarity is indeed complementary to gradient information.
  • HOG+HOF improves even more over HOG, especially for low false positive rates: at 0.1 fppi the improvement is 10.9%. This confirms previous results on the power of motion as a detection cue. Finally, HOG+HOF+CSS is better than only HOG+HOF, showing that CSS also contains information complementary to the flow, and achieves our best result of 44.35% recall at 0.1 fppi.
  • HOG+CSS provided better performance between 0.01 and 0.5 fppi, 6% at 0.1 fppi.
  • HOGF HOG
  • Adding HOF to HOG (HOGF) added 19.9% recall at 0.01 fppi.
  • HOG+CSS it beat the closest competitor HOG+CSS by 11% and the best published result (LatSvm ⁇ V2) by 21.2%.
  • Adding CSS brought another small improvement for large pedestrians.
  • the reason that HOF works so well on the “near” scale is probably that during multiscale flow estimation compression artifacts are less visible at higher pyramid levels, so that the flow field is more accurate for larger people.
  • Haar features did provide an improvement only on that dataset, on others they often cost performance. This is in contrast to CSS, which so far have produced consistent improvements, even on datasets with very different image quality and colour statistics. Judging from the available research, Haar features can potentially harm more than they help.
  • HOG+CSS consistently outperformed the results of [27] by 5%-8% against HOG+Haar with MPLBoost, and by 7%-8% against HOG with HIKSVM.
  • the detector of [27] using HOG+HOF in the IMHwd scheme
  • Haar features and a linear SVM is on par with HOG+HOF+CSS for low false positive rates, but it starts to fall back at 0.2 fppi.
  • the result of [27] using HOG+HOF with HIKSVM is consistently worse by 3%-5% than HOG+HOF+CSS, especially at low false positive rates.
  • Some of the method steps discussed above for determining a distribution density or determining self-similarity information, or detecting a human in the image for example, may be implemented by logic in the form of hardware or, for example, in software using a processing engine such as a microprocessor or a programmable logic device (PLD's) such as a PLA (programmable logic array), PAL (programmable array logic), FPGA (field programmable gate array).
  • a processing engine such as a microprocessor or a programmable logic device (PLD's) such as a PLA (programmable logic array), PAL (programmable array logic), FPGA (field programmable gate array).
  • PLA programmable logic array
  • PAL programmable array logic
  • FPGA field programmable gate array
  • This circuit may be constructed as a VLSI chip around an embedded microprocessor such as an ARM7TDMI core designed by ARM Ltd., UK which may be synthesized onto a single chip with the other components shown.
  • ARM7TDMI core designed by ARM Ltd., UK
  • other suitable processors may be used and these need not be embedded, e.g. a Pentium processor as supplied by Intel Corp. USA.
  • a zero wait state SRAM memory may be provided on-chip as well as a cache memory for example.
  • I/O (input/output) interfaces are provided for receiving and transmitting data to relevant networks, e.g. wireless or cable networks.
  • FIFO buffers may be used to decouple the processor from data transfer through these interfaces.
  • the interface can provide network connections, i.e. suitable ports and network addresses, e.g. the interfaces may be in the form of network cards.
  • Software programs may be stored in an internal ROM (read only memory) and/or on any other non-volatile memory, e.g. they may be stored in an external memory. Access to an external memory may be provided an external bus interface if needed, with address, data and control busses.
  • the method and apparatus of the embodiments described may be implemented as software to run on a processor.
  • an image processor in accordance with the present invention may be implemented by suitable programming of a processor.
  • the methods and procedures described above may be written as computer programs in a suitable computer language such as C and then compiled for the specific processor in the embedded design. For example, for the embedded ARM core VLSI described above the software may be written in C and then compiled using the ARM C compiler and the ARM assembler.
  • the software has code, which when executed on a processing engine provides the methods and the apparatus of the present invention.
  • the software programs may be stored on any suitable machine readable medium such as magnetic disks, diskettes, solid state memory, tape memory, optical disks such as CD-ROM or DVD-ROM, etc.
  • an image processor ( 10 ) has a window selector for choosing a detection window within the image, and a self similarity computation part ( 40 ) for determining self-similarity information for a group of the pixels in any part of the detection window, to represent an amount of self-similarity of that group to other groups in any other part of the detector window, and for repeating the determination for groups in all parts of the detection window, to generate a global self similarity descriptor for the detection window.
  • a classifier ( 50 ) is used for classifying whether an object is present based on the global self-similarity descriptor. By using global self-similarity rather than local similarities more information is captured which can lead to better classification. In particular, it helps enable recognition of more distant self-similarities inherent in the object, and self-similarities present at any scale. Other variations can be envisaged within the scope of the claims.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Human Computer Interaction (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Geometry (AREA)
  • Image Analysis (AREA)
US13/697,212 2010-06-11 2011-02-28 Detection of objects in an image using self similarities Abandoned US20130058535A1 (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
EP10165769.0 2010-06-11
EP10165769A EP2395452A1 (fr) 2010-06-11 2010-06-11 Détection d'objets dans une image en utilisant des autosimilarités
PCT/EP2011/052944 WO2011154166A1 (fr) 2010-06-11 2011-02-28 Détection d'objets dans une image à l'aide d'autosimilarités

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2011/052944 A-371-Of-International WO2011154166A1 (fr) 2010-06-11 2011-02-28 Détection d'objets dans une image à l'aide d'autosimilarités

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US14/989,504 Continuation US9569694B2 (en) 2010-06-11 2016-01-06 Detection of objects in an image using self similarities

Publications (1)

Publication Number Publication Date
US20130058535A1 true US20130058535A1 (en) 2013-03-07

Family

ID=42984005

Family Applications (2)

Application Number Title Priority Date Filing Date
US13/697,212 Abandoned US20130058535A1 (en) 2010-06-11 2011-02-28 Detection of objects in an image using self similarities
US14/989,504 Expired - Fee Related US9569694B2 (en) 2010-06-11 2016-01-06 Detection of objects in an image using self similarities

Family Applications After (1)

Application Number Title Priority Date Filing Date
US14/989,504 Expired - Fee Related US9569694B2 (en) 2010-06-11 2016-01-06 Detection of objects in an image using self similarities

Country Status (5)

Country Link
US (2) US20130058535A1 (fr)
EP (1) EP2395452A1 (fr)
JP (1) JP2013533998A (fr)
DE (1) DE112011101939T5 (fr)
WO (1) WO2011154166A1 (fr)

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140143183A1 (en) * 2012-11-21 2014-05-22 Disney Enterprises, Inc., A Delaware Corporation Hierarchical model for human activity recognition
US20150139490A1 (en) * 2012-06-14 2015-05-21 Katholieke Universiteit Leuven Discrimination container generation device and pattern detection device
US20150235139A1 (en) * 2014-02-20 2015-08-20 Nec Laboratories America, Inc. Extracting interpretable features for classification of multivariate time series from physical systems
US20160027180A1 (en) * 2014-07-25 2016-01-28 Samsung Electronics Co., Ltd. Method for retrieving image and electronic device thereof
US20160110882A1 (en) * 2013-06-25 2016-04-21 Chung-Ang University Industry-Academy Cooperation Foundation Apparatus and method for detecting multiple objects using adaptive block partitioning
US9380312B2 (en) 2014-07-14 2016-06-28 Apple Inc. Encoding blocks in video frames containing text using histograms of gradients
US9424657B2 (en) 2013-03-08 2016-08-23 Mstar Semiconductor, Inc. Image motion detection method, image processing method and apparatus using the methods
US9767370B1 (en) * 2014-09-17 2017-09-19 Waymo Llc Construction object detection
US9805274B2 (en) 2016-02-03 2017-10-31 Honda Motor Co., Ltd. Partially occluded object detection using context and depth ordering
US9922398B1 (en) 2016-06-30 2018-03-20 Gopro, Inc. Systems and methods for generating stabilized visual content using spherical visual content
US10043552B1 (en) 2016-10-08 2018-08-07 Gopro, Inc. Systems and methods for providing thumbnails for video content
US10056115B2 (en) 2014-07-03 2018-08-21 Gopro, Inc. Automatic generation of video and directional audio from spherical content
US10169684B1 (en) * 2015-10-01 2019-01-01 Intellivision Technologies Corp. Methods and systems for recognizing objects based on one or more stored training images
US20190019307A1 (en) * 2017-07-11 2019-01-17 Commissariat A L'energie Atomique Et Aux Energies Alternatives Image processing method
US10198657B2 (en) * 2016-12-12 2019-02-05 National Chung Shan Institute Of Science And Technology All-weather thermal-image pedestrian detection method
CN109389048A (zh) * 2018-09-13 2019-02-26 华南理工大学 一种监控视频中的行人检测与跟踪方法
US10341564B1 (en) 2018-05-18 2019-07-02 Gopro, Inc. Systems and methods for stabilizing videos
US10380732B2 (en) * 2016-06-02 2019-08-13 Komatsu Industries Corporation Controller, forming machine, and control method
US10432864B1 (en) 2018-09-19 2019-10-01 Gopro, Inc. Systems and methods for stabilizing videos
US10469818B1 (en) 2017-07-11 2019-11-05 Gopro, Inc. Systems and methods for facilitating consumption of video content
US10497121B2 (en) * 2014-10-27 2019-12-03 Alibaba Group Holding Limited Method and system for extracting a main subject of an image
US10684679B1 (en) 2016-10-21 2020-06-16 Gopro, Inc. Systems and methods for generating viewpoints for visual content based on gaze
CN112465028A (zh) * 2020-11-27 2021-03-09 南京邮电大学 一种感知视觉安全评估方法及系统
US11042775B1 (en) * 2013-02-08 2021-06-22 Brain Corporation Apparatus and methods for temporal proximity detection
CN113034454A (zh) * 2021-03-16 2021-06-25 上海交通大学 一种基于人类视觉感官的水下图像质量评价方法
US11170226B2 (en) * 2017-07-06 2021-11-09 Toyota Motor Europe Tracking objects in sequences of digital images
US20210365490A1 (en) * 2013-06-27 2021-11-25 Kodak Alaris Inc. Method for ranking and selecting events in media collections
US11195031B2 (en) 2016-12-16 2021-12-07 ZF Automotive UK Limited Method of determining the boundary of a driveable space
US11373439B1 (en) * 2013-03-14 2022-06-28 Telos Corporation Touchless fingerprint matching systems and methods

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5845139B2 (ja) * 2012-05-28 2016-01-20 日本電信電話株式会社 図形検出処理装置、図形検出処理方法及び図形検出処理プログラム
CN102831295B (zh) * 2012-07-19 2015-12-16 华北电力大学 基于纹理聚类的电力塔杆图像描述方法
US9846707B2 (en) 2013-01-31 2017-12-19 Bluebeam, Inc. Method for color and size based pre-filtering for visual object searching of documents
JP6194260B2 (ja) * 2014-02-21 2017-09-06 日本電信電話株式会社 画像分類装置、画像分類方法及び画像分類プログラム
CN106803056B (zh) * 2015-11-26 2020-11-06 华为技术有限公司 一种肢体关系的估计方法及装置
CN107798272B (zh) * 2016-08-30 2021-11-02 佳能株式会社 快速多目标检测与跟踪系统
TWI628623B (zh) * 2016-11-25 2018-07-01 國家中山科學研究院 全天候熱影像式行人偵測方法
CN107247936A (zh) * 2017-05-31 2017-10-13 北京小米移动软件有限公司 图像识别方法及装置
TWI651686B (zh) * 2017-11-30 2019-02-21 國家中山科學研究院 一種光學雷達行人偵測方法
CN109274538A (zh) * 2018-11-06 2019-01-25 四川长虹电器股份有限公司 一种基于fpga的自相似流量生成器实现方法
US20230081909A1 (en) * 2021-09-16 2023-03-16 Objectvideo Labs, Llc Training an object classifier with a known object in images of unknown objects
CN116805316B (zh) * 2023-08-25 2023-11-28 深圳市鹏顺兴包装制品有限公司 基于图像增强的可降解塑料加工质量检测方法

Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6373979B1 (en) * 1999-01-29 2002-04-16 Lg Electronics, Inc. System and method for determining a level of similarity among more than one image and a segmented data structure for enabling such determination
US20060029253A1 (en) * 2004-07-30 2006-02-09 Euclid Discoveries Llc Apparatus and method for processing video data
US20060228006A1 (en) * 2000-01-28 2006-10-12 Chuo Hatsujo Kabushiki Kaisha Fingerprint image evaluation method and fingerprint verification device
US20090208110A1 (en) * 2008-02-14 2009-08-20 Microsoft Corporation Factoring repeated content within and among images
US20100104158A1 (en) * 2006-12-21 2010-04-29 Eli Shechtman Method and apparatus for matching local self-similarities
US20100104203A1 (en) * 2002-02-13 2010-04-29 Reify Corporation Method and Apparatus for Acquisition, Compression, and Characterization of Spatiotemporal Signals
US20100191722A1 (en) * 2006-08-07 2010-07-29 Oren Boiman Data similarity and importance using local and global evidence scores
US20100188580A1 (en) * 2009-01-26 2010-07-29 Stavros Paschalakis Detection of similar video segments
US20100266212A1 (en) * 2009-04-20 2010-10-21 Ron Maurer Estimating Vanishing Points in Images
US20110075935A1 (en) * 2009-09-25 2011-03-31 Sony Corporation Method to measure local image similarity based on the l1 distance measure
US20120207396A1 (en) * 2011-02-15 2012-08-16 Sony Corporation Method to measure local image similarity and its application in image processing
US20120281923A1 (en) * 2011-05-02 2012-11-08 Yeda Research And Development Co., Ltd. Device, system, and method of image processing utilizing non-uniform image patch recurrence
US20130129143A1 (en) * 2011-11-21 2013-05-23 Seiko Epson Corporation Global Classifier with Local Adaption for Objection Detection
US20130156348A1 (en) * 2011-12-15 2013-06-20 Michal Irani Device, system, and method of visual inference by collaborative composition
US8615108B1 (en) * 2013-01-30 2013-12-24 Imimtek, Inc. Systems and methods for initializing motion tracking of human hands
US20150145862A1 (en) * 2013-11-27 2015-05-28 Adobe Systems Incorporated Texture Modeling of Image Data

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3836814B2 (ja) * 2003-05-20 2006-10-25 株式会社東芝 画像処理方法、画像処理装置、画像処理プログラム
JP2006268825A (ja) * 2005-02-28 2006-10-05 Toshiba Corp オブジェクト検出装置、学習装置、オブジェクト検出システム、方法、およびプログラム
US7899253B2 (en) * 2006-09-08 2011-03-01 Mitsubishi Electric Research Laboratories, Inc. Detecting moving objects in video by classifying on riemannian manifolds
JP2009193187A (ja) * 2008-02-13 2009-08-27 Casio Comput Co Ltd 画像検索方法、画像検索システム、画像検索端末および検索用サーバー
JP4997178B2 (ja) * 2008-06-10 2012-08-08 学校法人中部大学 物体検出装置
US9633275B2 (en) * 2008-09-11 2017-04-25 Wesley Kenneth Cobb Pixel-level based micro-feature extraction
EP2657857A1 (fr) * 2012-04-27 2013-10-30 ATG Advanced Swiss Technology Group AG Procédé de classification binaire d'une image de requête
US20140270489A1 (en) 2013-03-12 2014-09-18 Microsoft Corporation Learned mid-level representation for contour and object detection
CA2905637C (fr) * 2013-03-13 2022-04-05 Fdna Inc. Systemes, procedes, et supports lisibles par ordinateur pour identifier a quel moment un sujet est susceptible d'etre affecte par une condition medicale
US9098773B2 (en) * 2013-06-27 2015-08-04 Chevron U.S.A. Inc. System and method of detecting objects in scene point cloud
US20150185017A1 (en) * 2013-12-28 2015-07-02 Gregory L. Kreider Image-based geo-hunt
US9652851B2 (en) * 2014-04-01 2017-05-16 Conduent Business Services, Llc Side window detection in near-infrared images utilizing machine learning

Patent Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6373979B1 (en) * 1999-01-29 2002-04-16 Lg Electronics, Inc. System and method for determining a level of similarity among more than one image and a segmented data structure for enabling such determination
US20060228006A1 (en) * 2000-01-28 2006-10-12 Chuo Hatsujo Kabushiki Kaisha Fingerprint image evaluation method and fingerprint verification device
US20100104203A1 (en) * 2002-02-13 2010-04-29 Reify Corporation Method and Apparatus for Acquisition, Compression, and Characterization of Spatiotemporal Signals
US20060029253A1 (en) * 2004-07-30 2006-02-09 Euclid Discoveries Llc Apparatus and method for processing video data
US20100191722A1 (en) * 2006-08-07 2010-07-29 Oren Boiman Data similarity and importance using local and global evidence scores
US20100104158A1 (en) * 2006-12-21 2010-04-29 Eli Shechtman Method and apparatus for matching local self-similarities
US20090208110A1 (en) * 2008-02-14 2009-08-20 Microsoft Corporation Factoring repeated content within and among images
US20100188580A1 (en) * 2009-01-26 2010-07-29 Stavros Paschalakis Detection of similar video segments
US20100266212A1 (en) * 2009-04-20 2010-10-21 Ron Maurer Estimating Vanishing Points in Images
US20110075935A1 (en) * 2009-09-25 2011-03-31 Sony Corporation Method to measure local image similarity based on the l1 distance measure
US20120207396A1 (en) * 2011-02-15 2012-08-16 Sony Corporation Method to measure local image similarity and its application in image processing
US20120281923A1 (en) * 2011-05-02 2012-11-08 Yeda Research And Development Co., Ltd. Device, system, and method of image processing utilizing non-uniform image patch recurrence
US20130129143A1 (en) * 2011-11-21 2013-05-23 Seiko Epson Corporation Global Classifier with Local Adaption for Objection Detection
US20130156348A1 (en) * 2011-12-15 2013-06-20 Michal Irani Device, system, and method of visual inference by collaborative composition
US8615108B1 (en) * 2013-01-30 2013-12-24 Imimtek, Inc. Systems and methods for initializing motion tracking of human hands
US20150145862A1 (en) * 2013-11-27 2015-05-28 Adobe Systems Incorporated Texture Modeling of Image Data

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Dalal and Triggs. Histograms of Oriented Gradients for Human Detection. IEEE 2005, vol. 1, pgs. 886 - 893. *
Zhu et al. "Theory of Keyblock-Based Image Retrieval," ACM Transactions on Information Systems, Vol. 20, No. 2, April 2002, Pages 224-257. *

Cited By (57)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9818025B2 (en) * 2012-06-14 2017-11-14 Toyota Jidosha Kabushiki Kaisha Discrimination container generation device and pattern detection device
US20150139490A1 (en) * 2012-06-14 2015-05-21 Katholieke Universiteit Leuven Discrimination container generation device and pattern detection device
US20140143183A1 (en) * 2012-11-21 2014-05-22 Disney Enterprises, Inc., A Delaware Corporation Hierarchical model for human activity recognition
US9846845B2 (en) * 2012-11-21 2017-12-19 Disney Enterprises, Inc. Hierarchical model for human activity recognition
US11042775B1 (en) * 2013-02-08 2021-06-22 Brain Corporation Apparatus and methods for temporal proximity detection
US9424657B2 (en) 2013-03-08 2016-08-23 Mstar Semiconductor, Inc. Image motion detection method, image processing method and apparatus using the methods
TWI560649B (en) * 2013-03-08 2016-12-01 Mstar Semiconductor Inc Iamge motion detecting method, image processing method and apparatus utilizing these methods
US11373439B1 (en) * 2013-03-14 2022-06-28 Telos Corporation Touchless fingerprint matching systems and methods
US9836851B2 (en) * 2013-06-25 2017-12-05 Chung-Ang University Industry-Academy Cooperation Foundation Apparatus and method for detecting multiple objects using adaptive block partitioning
US20160110882A1 (en) * 2013-06-25 2016-04-21 Chung-Ang University Industry-Academy Cooperation Foundation Apparatus and method for detecting multiple objects using adaptive block partitioning
US20210365490A1 (en) * 2013-06-27 2021-11-25 Kodak Alaris Inc. Method for ranking and selecting events in media collections
US9652716B2 (en) * 2014-02-20 2017-05-16 Nec Corporation Extracting interpretable features for classification of multivariate time series from physical systems
US20150235139A1 (en) * 2014-02-20 2015-08-20 Nec Laboratories America, Inc. Extracting interpretable features for classification of multivariate time series from physical systems
US10056115B2 (en) 2014-07-03 2018-08-21 Gopro, Inc. Automatic generation of video and directional audio from spherical content
US10679676B2 (en) 2014-07-03 2020-06-09 Gopro, Inc. Automatic generation of video and directional audio from spherical content
US10410680B2 (en) 2014-07-03 2019-09-10 Gopro, Inc. Automatic generation of video and directional audio from spherical content
US10573351B2 (en) 2014-07-03 2020-02-25 Gopro, Inc. Automatic generation of video and directional audio from spherical content
US9380312B2 (en) 2014-07-14 2016-06-28 Apple Inc. Encoding blocks in video frames containing text using histograms of gradients
US20160027180A1 (en) * 2014-07-25 2016-01-28 Samsung Electronics Co., Ltd. Method for retrieving image and electronic device thereof
US9886762B2 (en) * 2014-07-25 2018-02-06 Samsung Electronics Co., Ltd. Method for retrieving image and electronic device thereof
US9767370B1 (en) * 2014-09-17 2017-09-19 Waymo Llc Construction object detection
US10497121B2 (en) * 2014-10-27 2019-12-03 Alibaba Group Holding Limited Method and system for extracting a main subject of an image
US10169684B1 (en) * 2015-10-01 2019-01-01 Intellivision Technologies Corp. Methods and systems for recognizing objects based on one or more stored training images
CN108604405A (zh) * 2016-02-03 2018-09-28 本田技研工业株式会社 利用环境和深度次序探测局部被遮挡的物体
US9805274B2 (en) 2016-02-03 2017-10-31 Honda Motor Co., Ltd. Partially occluded object detection using context and depth ordering
US10380732B2 (en) * 2016-06-02 2019-08-13 Komatsu Industries Corporation Controller, forming machine, and control method
US9922398B1 (en) 2016-06-30 2018-03-20 Gopro, Inc. Systems and methods for generating stabilized visual content using spherical visual content
US10607313B2 (en) 2016-06-30 2020-03-31 Gopro, Inc. Systems and methods for generating stabilized visual content using spherical visual content
US10043552B1 (en) 2016-10-08 2018-08-07 Gopro, Inc. Systems and methods for providing thumbnails for video content
US11614800B2 (en) 2016-10-21 2023-03-28 Gopro, Inc. Systems and methods for generating viewpoints for visual content based on gaze
US11061474B2 (en) 2016-10-21 2021-07-13 Gopro, Inc. Systems and methods for generating viewpoints for visual content based on gaze
US10684679B1 (en) 2016-10-21 2020-06-16 Gopro, Inc. Systems and methods for generating viewpoints for visual content based on gaze
US10198657B2 (en) * 2016-12-12 2019-02-05 National Chung Shan Institute Of Science And Technology All-weather thermal-image pedestrian detection method
US11195031B2 (en) 2016-12-16 2021-12-07 ZF Automotive UK Limited Method of determining the boundary of a driveable space
US11170226B2 (en) * 2017-07-06 2021-11-09 Toyota Motor Europe Tracking objects in sequences of digital images
US10469818B1 (en) 2017-07-11 2019-11-05 Gopro, Inc. Systems and methods for facilitating consumption of video content
US10783658B2 (en) * 2017-07-11 2020-09-22 Commissariat à l'Energie Atomique et aux Energies Alternatives Image processing method
US20190019307A1 (en) * 2017-07-11 2019-01-17 Commissariat A L'energie Atomique Et Aux Energies Alternatives Image processing method
US10341564B1 (en) 2018-05-18 2019-07-02 Gopro, Inc. Systems and methods for stabilizing videos
US11696027B2 (en) 2018-05-18 2023-07-04 Gopro, Inc. Systems and methods for stabilizing videos
US10587808B2 (en) 2018-05-18 2020-03-10 Gopro, Inc. Systems and methods for stabilizing videos
US11025824B2 (en) 2018-05-18 2021-06-01 Gopro, Inc. Systems and methods for stabilizing videos
US10587807B2 (en) 2018-05-18 2020-03-10 Gopro, Inc. Systems and methods for stabilizing videos
US11363197B2 (en) 2018-05-18 2022-06-14 Gopro, Inc. Systems and methods for stabilizing videos
US10574894B2 (en) 2018-05-18 2020-02-25 Gopro, Inc. Systems and methods for stabilizing videos
CN109389048A (zh) * 2018-09-13 2019-02-26 华南理工大学 一种监控视频中的行人检测与跟踪方法
US11172130B2 (en) 2018-09-19 2021-11-09 Gopro, Inc. Systems and methods for stabilizing videos
US10432864B1 (en) 2018-09-19 2019-10-01 Gopro, Inc. Systems and methods for stabilizing videos
US10536643B1 (en) 2018-09-19 2020-01-14 Gopro, Inc. Systems and methods for stabilizing videos
US11228712B2 (en) 2018-09-19 2022-01-18 Gopro, Inc. Systems and methods for stabilizing videos
US10958840B2 (en) 2018-09-19 2021-03-23 Gopro, Inc. Systems and methods for stabilizing videos
US11647289B2 (en) 2018-09-19 2023-05-09 Gopro, Inc. Systems and methods for stabilizing videos
US11678053B2 (en) 2018-09-19 2023-06-13 Gopro, Inc. Systems and methods for stabilizing videos
US10750092B2 (en) 2018-09-19 2020-08-18 Gopro, Inc. Systems and methods for stabilizing videos
US11979662B2 (en) 2018-09-19 2024-05-07 Gopro, Inc. Systems and methods for stabilizing videos
CN112465028A (zh) * 2020-11-27 2021-03-09 南京邮电大学 一种感知视觉安全评估方法及系统
CN113034454A (zh) * 2021-03-16 2021-06-25 上海交通大学 一种基于人类视觉感官的水下图像质量评价方法

Also Published As

Publication number Publication date
DE112011101939T5 (de) 2013-10-10
WO2011154166A1 (fr) 2011-12-15
US9569694B2 (en) 2017-02-14
EP2395452A1 (fr) 2011-12-14
JP2013533998A (ja) 2013-08-29
US20160117571A1 (en) 2016-04-28

Similar Documents

Publication Publication Date Title
US9569694B2 (en) Detection of objects in an image using self similarities
Klaser et al. A spatio-temporal descriptor based on 3d-gradients
Walk et al. New features and insights for pedestrian detection
Hu et al. Fast detection of multiple objects in traffic scenes with a common detection framework
Wang et al. Boosting dense SIFT descriptors and shape contexts of face images for gender recognition
Dollár et al. Integral channel features.
Zeng et al. Robust head-shoulder detection by PCA-based multilevel HOG-LBP detector for people counting
Wojek et al. A performance evaluation of single and multi-feature people detection
Zhang et al. Pedestrian detection in infrared images based on local shape features
Wang et al. Improved human detection and classification in thermal images
Han et al. A two-stage approach to people and vehicle detection with hog-based svm
Chen et al. Fast human detection using a novel boosted cascading structure with meta stages
Paisitkriangkrai et al. Fast pedestrian detection using a cascade of boosted covariance features
Wang et al. Dense sift and gabor descriptors-based face representation with applications to gender recognition
Daniel Costea et al. Word channel based multiscale pedestrian detection without image resizing and using only one classifier
Yao et al. Fast human detection from videos using covariance features
Cui et al. 3d haar-like features for pedestrian detection
JP2009510542A (ja) カメラによって取得されたシーンのテスト画像内の人物を検出する方法及びシステム
Chi Qin et al. An efficient method of HOG feature extraction using selective histogram bin and PCA feature reduction
CN114783003A (zh) 一种基于局部特征注意力的行人重识别方法和装置
Feichtenhofer et al. Dynamically encoded actions based on spacetime saliency
Yin et al. A face anti-spoofing method based on optical flow field
Liu et al. Related HOG features for human detection using cascaded adaboost and SVM classifiers
Corvee et al. Haar like and LBP based features for face, head and people detection in video sequences
Varga et al. Pedestrian detection in surveillance videos based on CS-LBP feature

Legal Events

Date Code Title Description
STCB Information on status: application discontinuation

Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION