Abstract: A spectrum filler for filling non-coded residual sub-vectors of a transform coded audio signal includes a sub-vector compressor configured to compress actually coded residual sub-vectors. A sub-vector rejecter is configured to reject compressed residual sub-vectors that do not fulfill a predetermined sparseness criterion. A sub-vector collector is configured to concatenate the remaining compressed residual sub-vectors to form a first virtual codebook. A coefficient combiner is configured to combine pairs of coefficients of the first virtual codebook to form a second virtual codebook. A sub-vector filler is configured to fill non-coded residual sub-vectors below a predetermined frequency with coefficients from the first virtual codebook, and to fill non-coded residual sub-vectors above the predetermined frequency with coefficients from the second virtual codebook.

Abstract: Techniques for distributed object detection and tracking are described. In an example method, a first current frame from a series of frames is sent to a first node, for detection of a first object. After object detection information for the first object is received in return, a second node is selected, and a second current frame is sent to the second node for an updated object detection. In addition, while waiting for the results of the updated object detection, two or more frames following the second current frame are sent to respective tracking nodes. Object modelling information indicating location and/or classification of one or more objects, as derived from the previously received object detection information is also sent to each of the respective tracking nodes. Tracking information for the first object is received from each of the respective tracking nodes.

Abstract: There is provided mechanisms for joint visual object detection and object mapping to a 3D model. A method is performed by an image processing device. The method includes obtaining a first sequence of digital images of a scene as captured by a first image capturing unit, and obtaining a second sequence of digital images of the scene as captured by a second image capturing unit. The second sequence of digital images is time-wise synchronized with the first sequence of digital images by being captured time-wise in parallel with the first sequence of digital images. The first image capturing unit has a narrower field of view than the field of view of the second image capturing unit. The first image capturing unit and the second image capturing unit have a known spatial relation. The method includes performing joint visual object detection and object mapping to the 3D model.

Abstract: A spectrum filler for filling non-coded residual sub-vectors of a transform coded audio signal includes a sub-vector compressor configured to compress actually coded residual sub-vectors. A sub-vector rejecter is configured to reject compressed residual sub-vectors that do not fulfill a predetermined sparseness criterion. A sub-vector collector is configured to concatenate the remaining compressed residual sub-vectors to form a first virtual codebook. A coefficient combiner is configured to combine pairs of coefficients of the first virtual codebook to form a second virtual codebook. A sub-vector filler is configured to fill non-coded residual sub-vectors below a predetermined frequency with coefficients from the first virtual codebook, and to fill non-coded residual sub-vectors above the predetermined frequency with coefficients from the second virtual codebook.

Abstract: AM and LM parameters to be used for adapting an ASR model are derived for each audio segment of an audio stream comprising multiple audio programs. A set of identifiers, including a speaker identifier, a speaker domain identifier and a program domain identifier, is obtained for each audio segment. The set of identifiers are used to select most suitable AM and LM parameters for the particular audio segment. The embodiments enable provision of maximum constraints on the AMs and LMs and enable adaptation of the ASR model on the fly for audio streams of multiple audio programs, such as broadcast audio. This means that the embodiments enable selecting AM and LM parameters that are most suitable in terms of ASR performance for each audio segment.

Abstract: The invention relates to a computer implemented method for associating objects in a video comprising subsequent frames, the method comprising obtaining first object proposal region information of a previous frame, determining second object proposal region information of a current frame, wherein the first and second object proposal region information are at least indicative of an appearance measure, a spatial location and a detection probability of each object proposal region of the respective frame, associating objects in the video by at least associating a first set of object proposal regions of the previous frame to a second set of object proposal regions of the current frame, wherein the object proposal regions are associated using distance measures calculated based on the appearance measures, the spatial locations and the detection probabilities.

Abstract: An object tracking, in particular adapted for real-time augmented reality applications, involves determining a location of an object (20) in a current frame (10) of a video stream (15), at a point in time following output of a preceding frame (11) of the video stream (15) but preceding output of the current frame (10), by starting from a location of the object (20) determined by an object-detection server (5) for a previous frame (12) of the video stream (15) and recursively track the location of the object (20) in frames (11) of the video stream (15) following the previous frame (12) up to the current frame (10) and recursively update a model of the object (20). Accurate objection detection from an object-detection server (5) can thereby be used even if the object was detected in a past frame (12) of the video stream (15) that has already been visualized.

Abstract: An encoder for encoding a parametric spectral representation (f) of auto-regressive coefficients that partially represent an audio signal. The encoder includes a low-frequency encoder configured to quantize elements of a part of the parametric spectral representation that correspond to a low-frequency part of the audio signal. It also includes a high-frequency encoder configured to encode a high-frequency part (fH) of the parametric spectral representation (f) by weighted averaging based on the quantized elements ({circumflex over (f)}L) flipped around a quantized mirroring frequency ({circumflex over (f)}m), which separates the low-frequency part from the high-frequency part, and a frequency grid determined from a frequency grid codebook in a closed-loop search procedure. Described are also a corresponding decoder, corresponding encoding/decoding methods and UEs including such an encoder/decoder.

Abstract: A spectrum filler for filling non-coded residual sub-vectors of a transform coded audio signal includes a sub-vector compressor configured to compress actually coded residual sub-vectors. A sub-vector rejecter is configured to reject compressed residual sub-vectors that do not fulfill a predetermined sparseness criterion. A sub-vector collector is configured to concatenate the remaining compressed residual sub-vectors to form a first virtual codebook. A coefficient combiner is configured to combine pairs of coefficients of the first virtual codebook to form a second virtual codebook. A sub-vector filler is configured to fill non-coded residual sub-vectors below a predetermined frequency with coefficients from the first virtual codebook, and to fill non-coded residual sub-vectors above the predetermined frequency with coefficients from the second virtual codebook.

Abstract: AM and LM parameters to be used for adapting an ASR model are derived for each audio segment of an audio stream comprising multiple audio programs. A set of identifiers, including a speaker identifier, a speaker domain identifier and a program domain identifier, is obtained for each audio segment. The set of identifiers are used to select most suitable AM and LM parameters for the particular audio segment. The embodiments enable provision of maximum constraints on the AMs and LMs and enable adaptation of the ASR model on the fly for audio streams of multiple audio programs, such as broadcast audio. This means that the embodiments enable selecting AM and LM parameters that are most suitable in terms of ASR performance for each audio segment.

Abstract: An encoder for encoding a parametric spectral representation (f) of auto-regressive coefficients that partially represent an audio signal. The encoder includes a low-frequency encoder configured to quantize elements of a part of the parametric spectral representation that correspond to a low-frequency part of the audio signal. It also includes a high-frequency encoder configured to encode a high-frequency part (fH) of the parametric spectral representation (f) by weighted averaging based on the quantized elements ({circumflex over (f)}L) flipped around a quantized mirroring frequency ({circumflex over (f)}m), which separates the low-frequency part from the high-frequency part, and a frequency grid determined from a frequency grid codebook in a closed-loop search procedure. Described are also a corresponding decoder, corresponding encoding/decoding methods and UEs including such an encoder/decoder.

Abstract: A spectrum filler for filling non-coded residual sub-vectors of a transform coded audio signal includes a sub-vector compressor configured to compress actually coded residual sub-vectors. A sub-vector rejecter is configured to reject compressed residual sub-vectors that do not fulfill a predetermined sparseness criterion. A sub-vector collector is configured to concatenate the remaining compressed residual sub-vectors to form a first virtual codebook. A coefficient combiner is configured to combine pairs of coefficients of the first virtual codebook to form a second virtual codebook. A sub-vector filler is configured to fill non-coded residual sub-vectors below a predetermined frequency with coefficients from the first virtual codebook, and to fill non-coded residual sub-vectors above the predetermined frequency with coefficients from the second virtual codebook.

Abstract: There is provided mechanisms for visual object detection in a sequence of images. A method is performed by a visual object detector (200). The method comprises obtaining (S102) a sequence of images of a scene. The sequence of images at least comprises a current image of the scene and a previous image of the scene. The method comprises extracting (S104) a set of objects from the sequence of images by performing visual object detection in the sequence of images. Performing the visual object detection in at least part of the current image is conditioned on a set of conditions being fulfilled. The set of conditions at least pertains to an image-wise descriptor classification score computed for at least one of the previous image and the current image and pertaining to which type of content the scene comprises, and an image overlapping score pertaining to how much overlap in image area there is between the previous image and the current image.

Abstract: A method performed by an image registration entity includes obtaining a matching between a first set of objects in a first image of a scene and a second set of objects in a second image of the scene, and obtaining a first set of key-points from the first image and a second set of key-points from the second image. Image registration is performed by matching the first set of key-points to the second set of key-points. Those of the first set of key-points that are mapped to objects in the first set of objects and that have matching objects in the second set of objects are restricted to only be matched to those of the second set of key-points that are mapped to any of the matching objects in the second set of objects. The matching between key-points resulting from the image registration is applied when constructing an image representation of the scene.

Abstract: AM and LM parameters to be used for adapting an ASR model are derived for each audio segment of an audio stream comprising multiple audio programs. A set of identifiers, including a speaker identifier, a speaker domain identifier and a program domain identifier, is obtained for each audio segment. The set of identifiers are used to select most suitable AM and LM parameters for the particular audio segment. The embodiments enable provision of maximum constraints on the AMs and LMs and enable adaptation of the ASR model on the fly for audio streams of multiple audio programs, such as broadcast audio. This means that the embodiments enable selecting AM and LM parameters that are most suitable in terms of ASR performance for each audio segment.

Abstract: Techniques for distributed object detection and tracking are described. In an example method, a first current frame from a series of frames is sent to a first node, for detection of a first object. After object detection information for the first object is received in return, a second node is selected, and a second current frame is sent to the second node for an updated object detection. In addition, while waiting for the results of the updated object detection, two or more frames following the second current frame are sent to respective tracking nodes. Object modelling information indicating location and/or classification of one or more objects, as derived from the previously received object detection information is also sent to each of the respective tracking nodes. Tracking information for the first object is received from each of the respective tracking nodes.

Abstract: The invention relates to a computer implemented method for associating objects in a video comprising subsequent frames, the method comprising obtaining first object proposal region information of a previous frame, determining second object proposal region information of a current frame, wherein the first and second object proposal region information are at least indicative of an appearance measure, a spatial location and a detection probability of each object proposal region of the respective frame, associating objects in the video by at least associating a first set of object proposal regions of the previous frame to a second set of object proposal regions of the current frame, wherein the object proposal regions are associated using distance measures calculated based on the appearance measures, the spatial locations and the detection probabilities.

Abstract: A method of providing an indication of objects in frames of a video segment and a video processor configured to perform such a method are described. First and second frames from the video segment are analyzed and candidate objects in the frames are determined. First and second sets of candidate objects in the frames are created that provide a size and location of each candidate object along with a probability that the determination as a candidate object is correct. Candidate objects that are determined to be the same object present in both the first and second frames are found. If the probability that the determination as a candidate object is correct from the second set is lower than the probability for the corresponding candidate object in the first set, the second set is amended by increasing the probability for the candidate object.

Abstract: A spectrum filler for filling non-coded residual sub-vectors of a transform coded audio signal includes a sub-vector compressor configured to compress actually coded residual sub-vectors. A sub-vector rejecter is configured to reject compressed residual sub-vectors that do not fulfill a predetermined sparseness criterion. A sub-vector collector is configured to concatenate the remaining compressed residual sub-vectors to form a first virtual codebook. A coefficient combiner is configured to combine pairs of coefficients of the first virtual codebook to form a second virtual codebook. A sub-vector filler is configured to fill non-coded residual sub-vectors below a predetermined frequency with coefficients from the first virtual codebook, and to fill non-coded residual sub-vectors above the predetermined frequency with coefficients from the second virtual codebook.

Abstract: An object tracking, in particular adapted for real-time augmented reality applications, involves determining a location of an object (20) in a current frame (10) of a video stream (15), at a point in time following output of a preceding frame (11) of the video stream (15) but preceding output of the current frame (10), by starting from a location of the object (20) determined by an object-detection server (5) for a previous frame (12) of the video stream (15) and recursively track the location of the object (20) in frames (11) of the video stream (15) following the previous frame (12) up to the current frame (10) and recursively update a model of the object (20). Accurate objection detection from an object-detection server (5) can thereby be used even if the object was detected in a past frame (12) of the video stream (15) that has already been visualized.