Systems, methods, tangible non-transitory computer-readable media, and devices for autonomous vehicle operation are provided. For example, a computing system can receive object data that includes portions of sensor data. The computing system can determine, in a first stage of a multiple stage classification using hardware components, one or more first stage characteristics of the portions of sensor data based on a first machine-learned model. In a second stage of the multiple stage classification, the computing system can determine second stage characteristics of the portions of sensor data based on a second machine-learned model. The computing system can generate an object output based on the first stage characteristics and the second stage characteristics. The object output can include indications associated with detection of objects in the portions of sensor data.

A method of detecting target object includes: extracting, through a neural network, a feature of a reference frame and a feature of a frame under detection; inputting each of at least two feature groups from at least two network layers of the neural network into a detector so as to obtain a corresponding detection result group output from the detector; wherein each feature group includes features of the reference frame and of the frame under detection, each detection result group includes a classification result and a regression result with respect to each of a plurality of candidate boxes for a feature group; and acquiring a bounding box for the target object in the frame under detection according to the at least two detection result groups.

An embodiment of the present disclosure relates to an image generation device including an image acquisition unit, a position recognition unit, a boundary line setting unit, and an image compositing unit. The image acquisition unit is mounted to a vehicle, and configured to acquire a plurality of captured images in which the periphery of the vehicle is imaged by a plurality of imaging units having an overlap region where imaging regions thereof partially overlap with each other. The position recognition unit is configured to recognize positions of specific obstacles which are one or more obstacles positioned in the overlap region. The boundary line setting unit is configured to set the boundary line so that each of the specific obstacles is contained in the image captured by the imaging unit which is closest in distance to the specific obstacle, among the plurality of imaging units. The image compositing unit is configured to generate a composite image in which the plurality of captured images are combined, using the boundary line as a boundary when combining the plurality of captured images.

The present invention relates to an image processing apparatus and a mobile robot including the same. The image processing apparatus according to an embodiment of the present invention includes an image acquisition unit for obtaining an image and a processor for performing signal processing on the image from the image acquisition unit, and the processor is configured to group super pixels in the image on the basis of colors or luminances of the image, calculate representative values of the super pixels and perform segmentation on the basis of the representative values of the super pixels. Accordingly, image segmentation can be performed rapidly and accurately.

An approach is provided for a redundant feature detection engine. The approach, for instance, involves segmenting an input image into a plurality of grid cells for processing by the redundant feature detection engine. The redundant feature detection engine includes a neural network. The approach also involves, for each of the plurality of grid cells, initiating a prediction of an object code by the redundant feature detection engine. The object code is a predicted feature that uniquely identifies an object depicted in the input image. The approach further involves aggregating the plurality of grid cells into one or more clusters based on the object code predicted for said each grid cell. The approach further involves predicting one or more features of the object corresponding to a respective cluster of the one or more clusters by merging one or more feature prediction outputs of said each grid cell in the respective cluster.

Controlling an unmanned aerial vehicle to traverse a portion of an operational environment of the unmanned aerial vehicle may include obtaining an object detection type, obtaining object detection input data, obtaining relative object orientation data based on the object detection type and the object detection input data, and performing a collision avoidance operation based on the relative object orientation data. The object detection type may be monocular object detection, which may include obtaining the relative object orientation data by obtaining motion data indicating a change of spatial location for the unmanned aerial vehicle between obtaining the first image and obtaining the second image based on searching along epipolar lines to obtain optical flow data.

Embodiments include apparatus and methods for automatic detection of pole-like objects for a location at a region of a roadway and automatic localization based on the detected pole-like objects. Pole-like objects are modeled as cylinders and the models are generated based on detected vertical clusters of point cloud data associated to corresponding regions along the region of the roadway. The modeled pole-like objects are stored in a database and associated with the region of the roadway. Sensor data from a user located at the region of the roadway is received. The pole-like object model is accessed and compared to the received sensor data. Based on the comparison, localization of the user located at the region of the roadway is performed.

An object recognition method uses a sensor configured to acquire a position of an object existing in a surrounding environment as point clouds including a plurality of detection points in a top view. The method includes: grouping point clouds according to a proximity; determining, when performing polygon approximation on the grouped point clouds, whether at least part of the detection points constituting the grouped point clouds are located in a blind zone of an approximate polygon acquired by the polygon approximation with respect to the sensor; recognizing the grouped point clouds as point clouds corresponding to plural objects when it is determined that the detection points are located in the blind zone with respect to the sensor; and recognizing the grouped point clouds as point clouds corresponding to a single object of the approximate polygon when it is determined that the detection points are not located in the blind zone.

Provided herein is a system and method that acquires data and determines a driving action based on the data. The system comprises a sensor, one or more processors, and a memory storing instructions that, when executed by the one or more processors, causes the system to perform, determining data of interest comprising an object, feature, or region of interest, determining whether a sharpness of the data of interest exceeds a threshold, in response to determining that the sharpness does not exceed a threshold, operating the sensor to increase the sharpness of the data of interest until the sharpness exceeds the threshold, in response to the sharpness exceeding the threshold, determining a driving action of a vehicle based on the data of interest, and performing the driving action

A self-propelled work machine that conducts work in a work area, comprising an imaging unit for capturing an image of a surrounding environment and a processing unit, wherein the processing unit performs a first process of detecting a predetermined object based on an imaging result, a second process of determining a degree of a backlight state in the imaging result, and a third process of conducting travel control of the work machine for any of making a turn, stopping, and moving backward in front of the detected object, and in a case where it is determined in the second process that the imaging result is in the backlight state, the processing unit sets a distance from the detected object when conducting the travel control in the third process to the work machine to be larger.