Abstract: Techniques are provided for generation of secure video and tamper detection of the secure video. A methodology implementing the techniques according to an embodiment includes selecting a subset of macroblocks from a video frame to be transmitted and calculating a low frequency metric on each of the selected macroblocks. The method also includes performing a hash calculation on the low frequency metrics to generate a frame signature; encrypting the frame signature (using a private key) to generate an encrypted watermark; and modifying pixels of each of the selected macroblocks to generate the secured video frame, the modifications based on bits of the encrypted watermark that are associated with the selected macroblock. The method further includes authenticating a received video frame by comparing a calculated frame signature to an authenticated frame signature, the authenticated frame signature decrypted (using a public key) from an extracted watermark of the received video frame.

Abstract: Techniques are provided for audio capture path evaluation of microphones incorporated into a device under test (DUT). A methodology implementing the techniques according to an embodiment includes estimating impulse responses (IRs) of the DUT microphones based on a comparison of a test audio signal received through the DUT microphones, at a selected measurement angle, to the test audio signal received through a reference microphone. The method also includes calculating group delays for the DUT microphones based on phase responses of the estimated IRs and calculating an average of the group delays. The method further includes calculating a distance, projected onto the measurement angle, between the DUT microphones and a geometric center of the DUT microphones. The distance is calculated as a product of the speed of sound and a difference between the average delay and the group delays for the DUT microphones. The process is repeated for additional measurement angles.

Abstract: Techniques are provided for unsupervised training of a neural network to perform compression of a high dynamic range (HDR) image. A methodology implementing the techniques according to an embodiment includes performing global tone mapping on an HDR training image to generate a low dynamic range (LDR) training image. The method also includes applying the neural network to the HDR training image and the LDR training image to generate a delta image representing image detail lost in the global tone mapping operation. The method further includes summing the delta image with the LDR training image to generate an output training image, and generating a loss function calculated from a weighted sum of a contrast loss and a compression loss. The contrast loss is based on the output training image and the HDR training image, and the compression loss is based on the output training image and the LDR training image.

Abstract: Techniques are disclosed to control a camera device such that memory contention and power consumption is reduced during video processing routines, generally referred to herein as media tasks. In particular, a workload scheduler is implemented in a camera HAL and is configured to dispatch captured image frames in an alternating manner between competing media tasks such that the processing of those image frames is performed sequentially, and thus, eliminates or otherwise mitigates memory contention. To this end, techniques variously disclosed herein can be used to enable low-cost, low-memory configured devices to perform concurrent media tasks on captured high-definition video at high framerates, without an undesirable decrease in performance and an increase in power consumption.

Abstract: Techniques are provided for processing video frames in a process flow that includes first and second computation engines. In an example, the first engine is an artificial intelligence based computation engine, and the second engine is a heuristics-based computation engine. A sequence of frames of a video includes a first and second frames that are two consecutive frames in the sequence. An analyzer determines whether the second frame has non-redundant information relative to the first frame. In response to the determination, the analyzer selects one of the first or second engine for processing at least a section of the second frame. For example, if the second frame has non-redundant information relative to the first frame, at least the section of the second frame is processed by the first engine. If the second frame does not include non-redundant information, the second frame is processed by the second engine.

Abstract: Techniques are provided for detection of laser-based audio injection attacks. A methodology implementing the techniques according to an embodiment includes calculating cross correlations between signals received from microphones of an array of two or more microphones. The method also includes identifying time delays associated with peaks of the cross correlations, and magnitudes associated with the peaks of the cross correlations. The method further includes calculating a time alignment metric based on the time delays and calculating a similarity metric based on the magnitudes. The method further includes generating a first attack indicator based on a comparison of the time alignment metric to a first threshold and generating a second attack indicator based on a comparison of the similarity metric to a second threshold. The method further includes providing warning of a laser-based audio attack based on the first attack indicator and/or the second attack indicator.

Abstract: Techniques are provided for audio capture path evaluation of microphones incorporated into a device under test (DUT). A methodology implementing the techniques according to an embodiment includes estimating impulse responses (IRs) of the DUT microphones based on a comparison of a test audio signal received through the DUT microphones, at a selected measurement angle, to the test audio signal received through a reference microphone. The method also includes calculating group delays for the DUT microphones based on phase responses of the estimated IRs and calculating an average of the group delays. The method further includes calculating a distance, projected onto the measurement angle, between the DUT microphones and a geometric center of the DUT microphones. The distance is calculated as a product of the speed of sound and a difference between the average delay and the group delays for the DUT microphones. The process is repeated for additional measurement angles.

Abstract: Techniques are provided for generation of secure video and tamper detection of the secure video. A methodology implementing the techniques according to an embodiment includes selecting a subset of macroblocks from a video frame to be transmitted and calculating a low frequency metric on each of the selected macroblocks. The method also includes performing a hash calculation on the low frequency metrics to generate a frame signature; encrypting the frame signature (using a private key) to generate an encrypted watermark; and modifying pixels of each of the selected macroblocks to generate the secured video frame, the modifications based on bits of the encrypted watermark that are associated with the selected macroblock. The method further includes authenticating a received video frame by comparing a calculated frame signature to an authenticated frame signature, the authenticated frame signature decrypted (using a public key) from an extracted watermark of the received video frame.

Abstract: Techniques are provided for detection of laser-based audio injection attacks through classification of the acoustic environment. A methodology implementing the techniques according to an embodiment includes broadcasting a reference signal over a loudspeaker into a local environment, and generating a reference model of the local environment based on analysis of a transformed version of that reference signal received through a microphone of the device. The method further includes generating an estimate model based on analysis of a segment of speech in an audio signal received through the microphone. The estimate model is associated with an environment in which the speech was generated. The method further includes calculating a similarity metric (e.g., mathematical distance) between the reference model and the estimate model, and providing warning of a laser-based audio attack if the similarity metric exceeds a threshold value associated with an attack.

Abstract: Techniques are provided for audio-based detection and tracking of an acoustic source. A methodology implementing the techniques according to an embodiment includes generating acoustic signal spectra from signals provided by a microphone array, and performing beamforming on the acoustic signal spectra to generate beam signal spectra, using time-frequency masks to reduce noise. The method also includes detecting, by a deep neural network (DNN) classifier, an acoustic event, associated with the acoustic source, in the beam signal spectra. The DNN is trained on acoustic features associated with the acoustic event. The method further includes performing pattern extraction, in response to the detection, to identify time-frequency bins of the acoustic signal spectra that are associated with the acoustic event, and estimating a motion direction of the source relative to the array of microphones based on Doppler frequency shift of the acoustic event calculated from the time-frequency bins of the extracted pattern.

Abstract: Techniques are provided for synthesis of transformed image views, based on a reference image, using depth information. The transformed image views may simulate a change in position or focal length of a camera that produced the reference image. An example system includes an image transformation circuit configured to transform the reference image corresponding to a first viewpoint, to a transformed image corresponding to a second viewpoint. The system also includes an inverse warping circuit configured to calculate a mapping from the pixels of the transformed image to corresponding pixels of the reference image. The system further includes a hole detection circuit configured to detect holes in the transformed image based on depth discontinuities between the reference and transformed images; and a hole filling circuit configured to in-fill the detected holes using a sampling of selected neighboring pixels from the reference image, to synthesize a view based on the transformed image.

Abstract: Techniques are disclosed to control a camera device such that memory contention and power consumption is reduced during video processing routines, generally referred to herein as media tasks. In particular, a workload scheduler is implemented in a camera HAL and is configured to dispatch captured image frames in an alternating manner between competing media tasks such that the processing of those image frames is performed sequentially, and thus, eliminates or otherwise mitigates memory contention. To this end, techniques variously disclosed herein can be used to enable low-cost, low-memory configured devices to perform concurrent media tasks on captured high-definition video at high framerates, without an undesirable decrease in performance and an increase in power consumption.

Abstract: Techniques are provided for multi-channel video encoding with cross-channel look ahead rate control. A methodology implementing the techniques according to an embodiment includes encoding a first video channel to generate a first output bitstream and first channel statistics including a quantization parameter, the first video channel comprising a first plurality of video frames at a first resolution. The method further includes encoding a second video channel to generate a second output bitstream encoded at a bit rate based on the first channel statistics, the second video channel comprising a second plurality of video frames at a second resolution. The second resolution may equal the first resolution depending on target bit rates for the channels. The method further includes performing look ahead processing on the first video channel to generate first channel look ahead statistics, and encoding the second video channel based on the first channel look ahead statistics.

Abstract: Techniques are provided for unsupervised training of a neural network to perform compression of a high dynamic range (HDR) image. A methodology implementing the techniques according to an embodiment includes performing global tone mapping on an HDR training image to generate a low dynamic range (LDR) training image. The method also includes applying the neural network to the HDR training image and the LDR training image to generate a delta image representing image detail lost in the global tone mapping operation. The method further includes summing the delta image with the LDR training image to generate an output training image, and generating a loss function calculated from a weighted sum of a contrast loss and a compression loss. The contrast loss is based on the output training image and the HDR training image, and the compression loss is based on the output training image and the LDR training image.

Abstract: Techniques are provided for segmentation of objects in video frames. A methodology implementing the techniques according to an embodiment includes receiving image frames, including an initial reference frame, and receiving a mask to outline a region in the reference frame that contains the object to be segmented. The method also includes calculating Gaussian mixture models associated with both the masked region and a background region external to the masked region. The method further includes segmenting the object from a current frame based on a modelling of the pixels within an active area of the current frame as a Markov Random Field of nodes for cost minimization. The costs are based in part on the Gaussian mixture models. The active area is based on the segmentation of a previous frame and on an estimation of optical flow between the previous frame and the current frame.

Abstract: An electronic assembly that includes a plurality of electronic substrates that are plugged into a polygonal shaped motherboard. The polygonal shape of the motherboard minimizes the electrical path between electronic substrates while providing enough space between adjacent substrates to allow a fluid to sufficiently remove heat generated by integrated circuits of the substrates.