Techniques are described for using computing devices to perform automated operations to generate mapping information using inter-connected images of a defined area, and for using the generated mapping information in further automated manners. In at least some situations, the defined area includes an interior of a multi-room building, and the generated information includes a floor map of the building, such as from an automated analysis of multiple panorama images or other images acquired at various viewing locations within the buildingin at least some such situations, the generating is further performed without having detailed information about distances from the images' viewing locations to walls or other objects in the surrounding building. The generated floor map and other mapping-related information may be used in various manners, including for controlling navigation of devices (e.g., autonomous vehicles), for display on one or more client devices in corresponding graphical user interfaces, etc.

A method includes receiving a user input (e.g., a one-touch user input), performing segmentation to generate multiple candidate regions of interest (ROIs) in response to the user input, and performing ROI fusion to generate a final ROI (e.g., for a computer vision application). In some cases, the segmentation may include motion-based segmentation, color-based segmentation, or a combination thereof. Further, in some cases, the ROI fusion may include intraframe (or spatial) ROI fusion, temporal ROI fusion, or a combination thereof.

A method and an apparatus for generating vehicle damage information are provided. The method includes: acquiring a damage area image of a target vehicle; performing image segmentation on the damage area image to obtain a first suspected damage area; inputting the damage area image to a pre-trained detection model to obtain a second suspected damage area, the detection model being configured to detect a location of the suspected damage area in the image; determining a damage image feature based on the first suspected damage area and the second suspected damage area; and inputting the damage image feature to a pre-trained classification model to generate a damage type, the classification model being configured to characterize a corresponding relationship between the image feature and the damage type.

Embodiments provide for perceptual data association received from at least a first and a second sensor disposed at different positions in an environment, in respective time series of local scene graphs that identify several characteristics of at least one object in the environment that are updated different rates; merging, at an output rate, characteristics for each given object from the several time series of local scene graphs that are updated at the output rate; merging, at the output rate, characteristics for each given object from the time series of local scene graphs that are updated at rates other than the output rate; and outputting, at the output rate, a time series of global scene graphs including the merged characteristics.

Previous self-driving car systems can detect objects separately with either vision systems, RADAR systems or LIDAR systems. In an embodiment of the present invention, an object fusion module normalizes sensor output from vision, RADAR, and LIDAR systems into a common format. Then, the system fuses the object-level sensor data across all systems by associating all objects detected and predicting tracks for all objects. The present system improves over previous systems by using the data from all sensors combined to develop a single set of knowledge about the objects around the self-driving car, instead of each sensor operating separately.

This document describes techniques for fine-motion virtual-reality or augmented-reality control using radar. These techniques enable small motions and displacements to be tracked, even in the millimeter or sub-millimeter scale, for user control actions even when those actions are small, fast, or obscured due to darkness or varying light. Further, these techniques enable fine resolution and real-time control, unlike conventional RF-tracking or optical-tracking techniques.

A computer-implemented method for processing a 3-D point cloud data and an associated image data to enrich the 3-D point cloud data with relevant portions of the image date. The method comprises generating a 3-D point cloud data tensor representative of information contained in the 3-D point cloud data and generating an image tensor representative of information contained in the image data; and then analyzing the image tensor to identify a relevant data portion of the image information relevant to the at least one object candidate. The method further includes amalgamating the 3-D point cloud data tensor with a relevant portion of the image tensor associated with the relevant data portion of the image information to generate an amalgamated tensor associated with the surrounding area and storing the amalgamated tensor to be used by a machine learning algorithm (MLA) to determine presence of the object in the surrounding area.

A method and device of multi-focal sensing of an obstacle. A method includes acquiring detection results of obstacles at multiple moments by utilizing a camera with long focus lens and a camera with short focus lens; performing a target tracking to the acquired detection results, to obtain at least two tracking sequences, wherein each tracking sequence includes detection results acquired at the multiple moments for a same obstacle; and matching two random tracking sequences of the at least two tracking sequences, and combining the two random tracking sequences into a combined tracking sequence, if the matching is successful.

A behavior understanding system and a behavior understanding method are provided. The behavior understanding system includes a sensor and a processor. The sensor senses a motion of a human body portion for a time period. A sequence of motion sensing data of the sensor is obtained. At least two comparing results respectively corresponding to at least two timepoints within the time period are generated according to the motion sensing data. The comparing result are generated through comparing the motion sensing data with base motion data. The base motion data is related to multiple base motions. A behavior information of the human body portion is determined according to the comparing results. The behavior information is related to a behavior formed by at least one of the base motions. Accordingly, the accuracy of behavior understanding can be improved, and the embodiments may predict the behavior quickly.

Vehicle perception techniques include obtaining a training dataset represented by N training histograms, in an image feature space, corresponding to N training images, K-means clustering the N training histograms to determine K clusters with respective K respective cluster centers, wherein K and N are integers greater than or equal to one and K is less than or equal to N, comparing the N training histograms to their respective K cluster centers to determine maximum in-class distances for each of K clusters, applying a deep neural network (DNN) to input images of the set of inputs to output detected/classified objects with respective confidence scores, obtaining adjusted confidence scores by adjusting the confidence scores output by the DNN based on distance ratios of (i) minimal distances of input histograms representing the input images to the K cluster centers and (ii) the respective maximum in-class.