Abstract: Presented are deep learning-based systems and methods for fusing sensor data, such as camera images, motion sensors (GPS/IMU), and a 3D semantic map to achieve robustness, real-time performance, and accuracy of camera localization and scene parsing useful for applications such as robotic navigation and augment reality. In embodiments, a unified framework accomplishes this by jointly using camera poses and scene semantics in training and testing. To evaluate the presented methods and systems, embodiments use a novel dataset that is created from real scenes and comprises dense 3D semantically labeled point clouds, ground truth camera poses obtained from high-accuracy motion sensors, and pixel-level semantic labels of video camera images. As demonstrated by experimental results, the presented systems and methods are mutually beneficial for both camera poses and scene semantics.

Abstract: In one embodiment, objects are tracked by calculating affinity scores with multiple cues in cascading stages. In each stage, affinity scores are calculated for currently detected objects based on a cue. The affinity scores are then fed into subsequent stages to inform subsequent calculations based on subsequent cues, where calculations can be skipped and resources saved. Object tracks are updated every frame, to assist in autonomous vehicle driving.

Abstract: The present application provides a mixed reality interaction method, an apparatus, and a storage medium, where the method includes obtaining position information of a real object in space in real time through a positioning chip disposed in the real object, and controlling, according to the position information of the real object, a virtual object in an augmented reality AR scenario to move in the AR scenario according to a moving track of the real object. The technical solution allows an object in the real world and an object in the virtual world seen by a user to be synchronized, thereby improving the user experience.

Abstract: Embodiments of the disclosure disclose a method for a vehicle avoiding an obstacle, an electronic device, and a computer storage medium, and relates to a field of unmanned driving technologies. The detailed implementation includes: obtaining size information of a vehicle; obtaining a safe distance in a traveling direction of the vehicle; determining an obstacle screening area of the vehicle at least based on the size information and the safe distance; and determining an obstacle avoiding path of the vehicle in response to determining that the obstacle is located in the obstacle screening area. With the technical solution of the disclosure, higher efficiency obstacle avoiding can be realized.

Abstract: A method and an apparatus for generating vehicle damage information are provided. The method includes: acquiring a to-be-processed vehicle image; for a target detection model in at least one pre-trained target detection model: inputting the vehicle image to the target detection model to generate a suspected damage area detection result; and determining a location of a suspected damage area in the vehicle image based on the generated suspected damage area detection result. A mechanism for detecting a suspected damage area is provided based on the target detection model, improving the vehicle damage assessment efficiency.

Abstract: The present disclosure discloses a method and an apparatus for recognizing an image. A specific implementation of the method comprises: acquiring a to-be-recognized image containing a to-be-recognized object; sending the to-be-recognized image to a server, and receiving identification information of a target object corresponding to the to-be-recognized object returned by the server, obtained by recognizing the to-be-recognized image, and a confidence parameter returned by the server; and determining the identification information of the target object as a recognition result when the confidence parameter is greater than a confidence threshold; or acquiring labeled information associated with the to-be-recognized image from a third-party platform and determining the labeled information as the recognition result when the confidence parameter is smaller than the confidence threshold.

Abstract: A method and apparatus for positioning a key point, a device, and a storage medium are provided. The method may include: extracting a first feature map and a second feature map of a to-be-positioned image, the first feature map and the second feature map being different feature maps; determining, based on the first feature map, an initial position of a key point in the to-be-positioned image; determining, based on the second feature map, an offset of the key point; and adding the initial position of the key point with the offset of the key point to obtain a final position of the key point.

Abstract: Embodiments of the present disclosure provide a method and a device for positioning a vehicle, a device, and a computer readable storage medium. The method includes: determining first edge information of a vehicle from an image related to the vehicle and acquired by a sensing device; determining a contour model corresponding to the vehicle; determining second edge information that matches the first edge information from an edge information set associated with the contour model, in which each edge information in the edge information set corresponds to a position of the contour model relative to the sensing device; and determining a position of the vehicle relative to the sensing device based on a position corresponding to the second edge information.

Abstract: Embodiments of the present disclosure relate to a method and apparatus for processing a video. The method may include: acquiring object regions obtained by performing object detection on a target video frame, a type of an object in each of the object regions being a preset type; determining, for an object region in the acquired object regions, in response to determining that the object region satisfies a preset condition, that the object region is a non-ROI; using an object region other than the non-ROI in the object regions of the target video frame as a ROI; and acquiring a quantization parameter change corresponding to each ROI, and encoding the target video frame based on the quantization parameter change.

Abstract: A method and an apparatus for recognizing a boundary of a traffic sign are provided. The method includes: acquiring at least one initial boundary point of a traffic sign in a to-be-processed image, the initial boundary point being used to mark a boundary of the traffic sign; dividing the to-be-processed image into at least one image segmentation area based on the at least one initial boundary point, the image segmentation area including at least one initial boundary point; importing, for an image segmentation area in the at least one image segmentation area, the image segmentation area into a pre-trained traffic sign boundary point recognition model to obtain an update boundary point corresponding to the image segmentation area; and determining a boundary point of the traffic sign based on all the update boundary point corresponding to the image segmentation area.

Abstract: A lane line tracking method and device are provided. The method can include: projecting a lane line detection result of previous K frames preceding a current frame in a camera coordinate system to a world coordinate system, to obtain a first projection result, wherein K is a positive integer greater than or equal to 1, and the previous K frames are consecutive K frames preceding the current frame; projecting the first projection result of the previous K frames in the world coordinate system to a camera coordinate system of the current frame, to obtain a second projection result; and determining, in the camera coordinate system of the current frame, lane line groups of the current frame, according to lane line groups in the second projection result of the previous K frames.

Abstract: The present disclosure provides a three-dimensional (3D) object detection method, a 3D object detection apparatus, an electronic device, and a readable storage medium, belonging to a field of computer vision technologies. Two-dimensional (2D) image parameters and initial 3D image parameters are determined for a target object. Candidate 3D image parameters are determined for the target object based on a disturbance range of 3D parameters and the initial 3D image parameters determined for the target object. Target 3D image parameters are selected for the target object from the candidate 3D image parameters determined for the target object based on the 2D image parameters. A 3D detection result of the target object is determined based on the target 3D image parameters.

Abstract: A positioning method includes acquiring an image of an area where a target object is located at a first time instant and multiple frames of point cloud data of an area where the target object is located at multiple time instants, wherein the multiple frames of point cloud data include first point cloud data of the area where the target object is located at the first time instant. The method also includes determining a point cloud map according to the multiple frames of point cloud data and acquiring a target feature vector according to the first point cloud data, the point cloud map and the image. The method further includes determining a positioning result of the target object according to the target feature vector.

Abstract: The present disclosure provides a method and apparatus for detecting a region of interest in a video, a device and a storage medium. The method may include: acquiring a current to-be-processed frame from a picture frame sequence of a video; detecting a region of interest (ROI) in the current to-be-processed frame, in response to determining that the current to-be-processed frame is a detection picture frame, to determine at least one ROI in the current to-be-processed frame; and updating a to-be-tracked ROI, based on the ROI in the current to-be-processed frame and a tracking result determined by a pre-order tracking picture frame; and tracking the current to-be-processed frame based on the existing to-be-tracked ROI, in response to determining that the current to-be-processed frame is a tracking picture frame, to determine at least one tracking result as the ROI of the current to-be-processed frame.

Abstract: Embodiments of the present disclosure disclose a method and apparatus for identifying a video. A specific embodiment of the method includes: acquiring a predetermined number of video frames from a video to be identified to obtain a video frame sequence; performing the following processing step: importing the video frame sequence into a pre-trained video identification model to obtain a classification tag probability corresponding to the video frame sequence, wherein the classification tag probability is used to characterize a probability of identifying a corresponding tag category of the video to be identified; and setting, in response to the classification tag probability being greater than or equal to a preset identification accuracy threshold, a video tag for the video to be identified according to the classification tag probability, or else increasing the number of video frames in the video frame sequence and continuing to perform the above processing step.

Abstract: A method for determining an object comprises: determining a reference representation based on positions of a first reference point and a second reference point of a first object in a reference image, which indicates a distance and a relative direction between the first and second reference points in the reference image; determining a detection representation based on positions of the first and second reference points in a detection frame including the first object in a video, which indicates a distance and a relative direction between the first and second reference points in the detection frame; determining rotation parameters associated with a first predetermined point and a second predetermined point in a second object associated with the first object based on the reference representation and the detection representation; and adjusting a target representation between the first and second predetermined points in the second object based on the rotation parameter.

Abstract: The present application provides a method, an apparatus, a device and a system for processing commodity identification and a storage medium, where the method includes: receiving image information transmitted by a camera apparatus and a distance signal transmitted by a distance sensor corresponding to the camera apparatus; determining a start frame and an end frame for a pickup behavior of a user according to the image information and the distance signal; and determining, according to the start frame and the end frame for the pickup behavior of the user, information of a commodity taken by the user. By performing a commodity identification on the start frame and the end frame for the pickup behavior of the user, and determining the information of the commodity taken by the user, commodity identification efficiency is effectively improved.

Abstract: An image identification method and an image identification device are provided. The method comprises acquiring a hand feature region within a sight from a first view by skin color detection; acquiring a feature and a position of a tip of a finger from the hand feature region by performing a pattern recognition for a morphological feature of a stretched hand; recording an interframe displacement of a feature point of the tip of the finger when the tip of the finger delimits a periphery of a target object to obtain a delimiting trajectory from the interframe displacement, closing the delimiting trajectory to form a full-perimeter geometry; projecting the full-perimeter geometry on a plane where a direction of the sight is perpendicular to a plane where the target object is located to obtain a projection region, performing an image identification using the projection region as an identification region of the target object.

Abstract: A method, an apparatus, a device and a storage medium for three-dimensional reconstruction are provided. The method may include: acquiring a sequence of RGBD images of a target; determining a set of first three-dimensional key points of the target on each RGBD image for the each RGBD image in the sequence of the RGBD images; transferring the set of the first three-dimensional key points to a camera coordinate system to obtain a set of second three-dimensional key points of the target in the camera coordinate system; calculating an Euler angle of the each RGBD image based on the set of the second three-dimensional key points; selecting an RGBD image based on the Euler angle; and performing three-dimensional reconstruction of the target based on the selected RGBD image to obtain a three-dimensional model of the target.

Abstract: A method, an apparatus, a device and a computer-readable storage medium for recognizing aerial handwriting are provided. The method may include detecting a palm region of a user in a two-dimensional grayscale image; detecting a fingertip in the two-dimensional gray-scale image based on the palm area; determining a spatial trajectory of the fingertip based on a set of two-dimensional gray-scale images, the set of two-dimensional gray-scale images including the two-dimensional gray-scale image; and recognizing handwritten content of the user based on the spatial trajectory. A two-dimensional gray-scale image is used to recognize and track the spatial trajectory of the fingertip, which may speed up aerial handwriting recognition, and has low processing performance requirements for the device, while also ensuring high accuracy.