Systems and methods are disclosed for tracking objects in a physical environment using visual sensors onboard an autonomous unmanned aerial vehicle (UAV). In certain embodiments, images of the physical environment captured by the onboard visual sensors are processed to extract semantic information about detected objects. Processing of the captured images may involve applying machine learning techniques such as a deep convolutional neural network to extract semantic cues regarding objects detected in the images. The object tracking can be utilized, for example, to facilitate autonomous navigation by the UAV or to generate and display augmentative information regarding tracked objects to users.

The present invention discloses a multispectral camera external parameter self-calibration algorithm based on edge features, and belongs to the field of image processing and computer vision. Because a visible light camera and an infrared camera belong to different modes, fewer satisfactory point pairs are obtained by directly extracting and matching feature points. In order to solve the problem, the method starts from the edge features, and finds an optimal corresponding position of an infrared image on a visible light image through edge extraction and matching. In this way, a search range is reduced and the number of the satisfactory matched point pairs is increased, thereby more effectively conducting joint self-calibration on the infrared camera and the visible light camera. The operation is simple and results are accurate.

A camera and a vehicle including the same are disclosed. The camera includes a first camera, an image sensor to sense an image based on light acquired by the first camera, and a processor to control the exposure time of the image sensed by the image sensor to be changed based on a brightness level and a brightness distribution value of the image acquired by the image sensor. Consequently, it is possible to reduce motion blur in a low luminance environment.

A multi-camera time slice system and method of producing a three-dimensional (3D) integrated time slice image array uses multiple cameras to capture a subject to produce an array of subject images with transparent backgrounds. The array of subject images with transparent backgrounds is merged with an array of foreground images with transparent backgrounds and an array of background images, which are captured in a similar manner as the array of subject images, to produce the 3D integrated time slice image array.

In the related art, it was difficult to compare lengths of a plurality of objects which were present at different places. However, it is possible to easily compare lengths of photographed objects using an image capturing device which displays a length of an object which is calculated based on parallax information, by obtaining an image in which the object is photographed and the parallax information corresponding to the image as inputs, the device including an object extraction unit which extracts an image of an object using the parallax information from the photographed image; a comparison data maintaining unit which maintains the image of the object and the length of the object; an object comparison unit which compares the length of the object which is extracted using the object extraction unit to a length of comparison data which is extracted from the comparison data maintaining unit; and an image composition unit which combines a comparison result with the photographed image, and outputs the image.

A capability for managing a representation of a smart environment is presented herein. The capability for managing a representation of a smart environment is configured to support augmented reality (AR)-based management of a representation of a smart environment, which may include AR-based generation of a representation of the smart environment, AR-based alignment of the representation of the smart environment with the physical reality of the smart environment, and the like.

An apparatus for reducing the visibility of disparity estimation errors at edges, and in particular at overlays. The apparatus comprises a receiver (401) for receiving a three dimensional image represented by at least image values (brightness/contrast values) and a disparity value. A subset selector (403) evaluates an image property criterion for the image value for a group of pixels and determines a subset of pixels of the group of pixels for which the image property criterion is met. The criterion may for example reflect whether the pixel belongs to an image object edge. A distribution evaluator (405) generates a frequency distribution for disparity values of the subset of pixels and an analyzer (407) determines a shape property for the frequency distribution (the presence of a peak). An adaptor (409) determining a disparity remapping in response to the shape property and a remapper (411) modifies disparity values of the three dimensional image by applying the disparity remapping. The approach may e.g. reduce image depth when overlay graphics is likely to be present.

A method and device for detecting a vehicle pose, relating to the fields of computer vision and automatic driving. The specific implementation solution comprises: inputting a vehicle left view point image and a vehicle right view point image into a part prediction and mask segmentation network model, and determining foreground pixel points and part coordinates thereof in a reference image; converting coordinates of the foreground pixels in the reference image into coordinates of the foreground pixels in a camera coordinate system so as to obtain a pseudo-point cloud, and fusing part coordinate of the foreground pixels and the pseudo-point cloud to obtain fused pseudo-point cloud; and inputting the fused pseudo-point cloud into a pre-trained pose prediction model to obtain a pose information of the vehicle to be detected.

Disclosed is a cross-view image optimizing method and apparatus, and a computer equipment and a readable storage medium. The method includes: acquiring a sample image and a pre-trained cross-view image generating model; generating an multi-dimensional cross-view image of the sample image by a multi-dimensional feature extracting module of the first generator to obtain dimension features and cross-view initial images at multiple dimensions; obtaining a multi-dimensional feature map with corresponding dimension features by the second generator; inputting the multi-dimensional feature map to a multi-channel attention module of the second generator for feature extraction and calculating a feature weight of each attention channel, obtaining attention feature images, attention images and feature weights in a preset number of the attention channels; and weighting and summing the attention images and the attention feature images of all the channels according to the feature weights, and obtaining a cross-view target image.

It is provided a method for detecting an object in a left view image and a right view image, comprising steps of receiving the left view image and the right view image; detecting a coarse region containing the object in one image of the left view image and the right view image; detecting the object within the detected coarse region in the one image; determining a coarse region in the other image of the left view image and the right view image based on the detected coarse region in the one image and offset relationship indicating position relationship of the object in a past left view image and a past right view image; and detecting the object within the determined coarse region in the other image.