Abstract: Methods, systems, and devices for change detection are described. The methods, systems, and devices relate to monitoring a field of view of an image sensor via a first pixel associated with a group of pixels of the image sensor, where a dimension of the first pixel exceeds a dimension of at least one pixel of the group of pixels of the image sensor, detecting a change in the field of view of the image sensor based on the monitoring, and activating a second pixel of the image sensor based on detecting the change in the field of view of the image sensor.

Abstract: Apparatuses, methods, and systems are presented for sensing scene-based occurrences. Such an apparatus may comprise a vision sensor system comprising a first processing unit and dedicated computer vision (CV) computation hardware configured to receive sensor data from at least one sensor array comprising a plurality of sensor pixels and capable of computing one or more CV features using readings from neighboring sensor pixels. The vision sensor system may be configured to send an event to be received by a second processing unit in response to processing of the one or more computed CV features by the first processing unit. The event may indicate possible presence of one or more irises within a scene.

Abstract: An apparatus may include an anti-color filter array proximate an array of optical sensor pixels. The anti-color filter array may include first anti-color filters. Each of the first anti-color filters may be located near a corresponding instance (or group of instances) of a first plurality of optical sensor pixels and may be configured to at least partially suppress transmission of light in a first optical wavelength range. In some examples, the first optical wavelength range may correspond to a first color. The anti-color filter array may include second anti-color filter regions, each of which may be located near a corresponding instance of a second plurality of optical sensor pixels. In some such examples, each of the second anti-color filter regions may be configured to at least partially suppress transmission of light in a second optical wavelength range.

Abstract: Aspects described herein provide a method including: receiving input data at a machine learning model, comprising: a plurality of processing layers; a plurality of gate logics; a plurality of gates; and a fully connected layer; determining based on a plurality of gate parameters associated with the plurality of gate logics, a subset of the plurality of processing layers with which to process the input data; processing the input data with the subset of the plurality of processing layers and the fully connected layer to generate an inference; determining a prediction loss based on the inference and a training label associated with the input data; determining an energy loss based on the subset of the plurality of processing layers used to process the input data; and optimizing the machine learning model based on: the prediction loss; the energy loss; and a prior probability associated with the training label.

Abstract: Sensing of scene-based occurrences is disclosed. In one example, a vision sensor system comprises (1) dedicated computer vision (CV) computation hardware configured to receive sensor data from at least one sensor array and capable of computing CV features using readings from multiple neighboring sensor pixels and (2) a first processing unit communicatively coupled with the dedicated CV computation hardware. The vision sensor system is configured to, in response to processing of the one or more computed CV features indicating a presence of one or more irises in a scene captured by the at least one sensor array, generate data in support of iris-related operations to be performed by a second processing unit and send the generated data to the second processing unit.

Abstract: Methods, systems, computer-readable media, and apparatuses for region-of-interest automatic gain or exposure control are presented. One example method includes receiving an image from an image sensor, the image comprising a plurality of pixels, determining a region of interest (“ROI”) in the image, the ROI comprising a subset of the plurality of pixels in the image, determining a ROI weight for pixels in the subset of the plurality of pixels, determining statistics for the image based on the plurality of pixels in the image and the ROI weight, and adjusting a gain or exposure setting for the image sensor based on the statistics.

Abstract: Methods, systems, and devices for configuring a machine learning network are described. A device, which may be otherwise known as user equipment (UE), may support ultra-low power sensor applications. More specifically, the device may support memory block allocation of a machine learning network based on performance levels associated with the applications. For example, the device may identify a performance level associated with an application on the device. The device may determine that the performance level satisfies a condition, and subsequently determine a memory block allocation of a machine learning network of the device based on the performance level satisfying the condition. The memory block allocation may correspond to one or more connections of the machine learning network. Based on the memory block allocation, the device may adjust a quantity of memory blocks available for the machine learning network and process the application.

Abstract: Methods, systems, and devices for change detection are described. The methods, systems, and devices relate to monitoring a field of view of an image sensor via a first pixel associated with a group of pixels of the image sensor, where a dimension of the first pixel exceeds a dimension of at least one pixel of the group of pixels of the image sensor, detecting a change in the field of view of the image sensor based on the monitoring, and activating a second pixel of the image sensor based on detecting the change in the field of view of the image sensor.

Abstract: Cameras with large field-of-view lenses can cause significant geometric distortions of the images acquired. Training for object detection normally takes place on undistorted images. Thus, in order to detect objects of interest within the acquired images, an undistortion procedure is applied on the acquired images and an object detection is then performed on the undistorted images. Unfortunately, such undistortion procedure is too computationally expensive to be run on some edge devices. To remove the need to perform the undistortion procedure, it is proposed to train for object detection directly from distorted acquired images.

Abstract: Disclosed is a method and apparatus for operating a first device. The first device obtains environment information in proximity to the first device and receives one or more communication messages from a second device and the message(s) includes relevance criteria, wherein the relevance criteria indicates one or more devices, one or more sets of device characteristics, one or more lanes, one or more intersections or areas, one or more pedestrian paths or bicycle paths, one or more signal characteristics from the second device or any combination thereof. The first device determines whether the one or more communication messages are relevant to the first device based on the relevance criteria and the environment information and performs an operation in response to the determination of whether the one or more communication messages are relevant.

Abstract: Apparatus, methods, systems, and instructions stored on computer-readable medium are presented for performing classification. A hardware engine may be configurable to implement a plurality of cascade classifiers comprising at least a first cascade classifier and a second cascade classifier, each including one or more stages. The first cascade classifier may be configured to perform a classification different from a classification performed by the second cascade classifier. The hardware engine may be capable of conditionally branching from the first cascade classifier to the second cascade classifier based on an outcome condition associated with evaluating of at least one stage of the first cascade classifier.

Abstract: Some dynamic pixel binning methods may involve identifying ROI optical sensor data corresponding to optical sensor pixels in a region of interest (ROI). Some such methods also may involve identifying non-ROI optical sensor data corresponding to optical sensor pixels outside the region of interest and determining a first binning factor for the non-ROI optical sensor data. The first binning factor may be based on a number of pixels in the region of interest and a target output data rate. Some such methods also may involve applying the first binning factor to the non-ROI optical sensor data and outputting optical sensor data, including the binned non-ROI optical sensor data, at the target output data rate.

Abstract: Various aspects of the present disclosure generally relate to optical character detection. In some aspects, a user device may receive, from a vision sensor, a first image that is associated with a first optical character image. The user device may determine, using an image processing model, that the first image depicts the first optical character image. The user device may cause, based at least in part on determining that the first image depicts the first optical character image, a camera to capture a second image that is associated with a second optical character image. The user device may perform an action associated with the second image. Numerous other aspects are provided.

Abstract: Internet of Things (IoT) systems and related methods are disclosed. A method comprises determining whether a first condition holds, the first condition being a condition of the IoT system, performing a function of the IoT system in response to a trigger, wherein the trigger is a determination that the first condition holds, recognizing a gesture based on image data received from an image sensor, and reconfiguring the IoT system for a future performance of the function based on the user feedback data.

Abstract: In one example, an image sensor module comprises one or more covers having at least a first opening and a second opening, a first lens mounted in the first opening and having a first field of view (FOV) centered at a first axis having a first orientation, a second lens mounted in the second opening and having a second FOV centered at a second axis having a second orientation different from the first orientation, a first image sensor housed within the one or more covers and configured to detect light via the first lens, and a second image sensor housed within the one or more covers and configured to detect light via the second lens. The first image sensor and the second image sensor are configured to provide, based on the detected light, image data of a combined FOV larger than each of the first FOV and the second FOV.

Abstract: Disclosed is a method and apparatus for generating a real-world traffic model. The apparatus obtains a first set of device map information associated with one or more devices that are in proximity with a first device, and obtains a second set of device map information associated with one or more devices that are in proximity with a second device. The apparatus determines whether the first set of device map information and the second set of device map information contain at least one common device and in response to the determination that the first set of device map information and the second set of device map information contain at least one common device, and generates a real-world traffic model of devices based on the first set of device map information and the second set of device map information.

Abstract: Disclosed embodiments pertain to a method for determining position information of a target vehicle relative to an ego vehicle. The method may comprise: obtaining, by at least one image sensor, first images of one or more target vehicles and classifying at least one target vehicle from the one or more target vehicles based on the one or more first images. Further, vehicle characteristics corresponding to the least one target vehicle may be obtained based on the classification of the least one target vehicle. Position information of the at least one target vehicle relative to the ego vehicle may be determined based on the vehicle characteristics.

Abstract: Methods, systems, and devices for exposure control are described, including capturing a first and second field of view with a first and second sensor. The techniques may include identifying a brightness difference and an exposure time difference between the first and second sensor, and capturing a first image and a second image, and outputting a third image including both the first and second image. Techniques may include determining an exposure bias, identifying a hypothesis total gain for the first sensor and a peer sensor total gain for the second sensor, and selecting a total gain for each sensor based on comparing the hypothesis total gain and the peer sensor total gain, and based on a maximum brightness difference between the two sensors. The total gain for each sensor may be adjusted to satisfy the maximum brightness difference and the exposure bias, or based on a region of interest.

Abstract: Embodiments described herein can address these and other issues by providing a CV image sensor that has a repeating pattern of two pairs of color-opponent color sensors and a luminance sensor to provide a high amount of functionality while maintaining relatively low power requirements. By using color opponency, embodiments can sidestep the need for processor-intensive color conversions required of other color sensors. Embodiments may optionally provide for surface area optimization, hyperspectral sensitivity, low-light optimization, neuromorphic light sensing, and more.

Abstract: Some dynamic pixel binning methods may involve identifying ROI optical sensor data corresponding to optical sensor pixels in a region of interest (ROI). Some such methods also may involve identifying non-ROI optical sensor data corresponding to optical sensor pixels outside the region of interest and determining a first binning factor for the non-ROI optical sensor data. The first binning factor may be based on a number of pixels in the region of interest and a target output data rate. Some such methods also may involve applying the first binning factor to the non-ROI optical sensor data and outputting optical sensor data, including the binned non-ROI optical sensor data, at the target output data rate.