A far-infrared image acquisition unit acquires a far-infrared image. An image conversion unit converts the acquired far-infrared image into a visible light image. A visible light image trained model storage unit stores a first visible light image trained model having performed learning using the visible light image as training data. A transfer learning unit performs transfer learning on a first visible light image trained model by using the visible light image obtained by conversion as training data to generate a second visible light image trained model.

A method and apparatus for forklift pickup, a computer device, and a storage medium are provided in the disclosure. The method includes the following. Observational data of a truck parking area is obtained by observing the truck parking area. Point cloud data of at least one truck is obtained in the observational data, and point cloud data of each of the at least one carrier is obtained in the point cloud data of the at least one truck. A relative pose of each of at least one carrier is determined based on the point cloud data of each of the at least one carrier. A pickup priority is determined based on the point cloud data of each of the at least one carrier. A forklift is controlled to perform pickup according to the relative pose of each of the at least one carrier and the pickup priority

The disclosed technology provides solutions for object detection and classification with both high recall and high precision, by using a first stage with high recall, and a second stage to provide high precision. The dimensional state of a pointcloud is reduced from 3 to 2, and proposed bounding boxes are generated. The original pointcloud data is filtered according to the bounding boxes, and fused with learned features, with the fused data processed to generate the high recall and high precision output.

Systems and methods for performing long-term kinematic tracking of an animal subject are provided. Chronically-affixed motion capture markers including a tissue engaging feature and a reflective marker are described, the motion capture markers enabling long-term motion capture recording of an animal subject. A method of determining a three-dimensional pose of a subject using a trained statistical model configured to generate landmark position data associated with the three-dimensional pose of the animal subject. The method includes using projective geometry to generate three-dimensional image volumes as input to the trained statistical model. Further, a method for profiling a subject’s physical behavior over a period of time by applying clustering to information indicative of movement of the subject over the period of time is described.

Disclosed is a system and method for segmentation of selected data. In various embodiments, automatic segmentation of fiber tracts in an image data may be performed. The automatic segmentation may allow for identification of specific fiber tracts in an image.

Methods, systems, and computer programs are presented for adding new features to a network service. An example method includes accessing an image from a user device to determine a salient object count of a plurality of objects in the image. A salient object count of the plurality of objects in the image is determined. An indicator of the salient object count of the plurality of objects in the image is caused to be displayed on the user device.

The invention relates to a method for detecting an object (303, 304) in the vicinity of a road user. The method recognizes objects (303, 304) detected by the sensor unit (101) with the object recognition unit (102) of a device (100), stores detected objects (303, 304) in an object list with the memory unit (103), transfers the object list with the transfer unit (104) to a data cloud (306), determines a position of the first road user, transmits the position of the first road user to the data cloud (306), identifies an object (303, 304) near the first traffic participant in the data cloud (306) using the object list, transmits a position and an object type of the identified object (303, 304) from the data cloud (306) to a mobile device of the first road user and displays the object (303, 304) and its position on the mobile device of the first road user.

A method, computer program product, and system include a processor(s) that engages, based on a request for an inference, from a group of sensors of multiple modalities at a physical location, sensor(s) of a main modality to provide data to a pipeline to generate the inference. The pipeline includes one or more machine learning models which generate the inference for a downstream task. The processor(s) obtains raw data from the sensor(s) of the main modality and applies an outlier detector to the raw data. Based on determining that there is an outlier the processor(s) automatically engages sensor(s) of at least one different modality than the main modality from the group of sensors of multiple modalities and obtains new raw data from the sensor(s) of the at least one different modality. The processor(s) applies the one or more machine learning models to the new raw data to derive the inference.

An object detection method includes steps that are to be performed for each piece of point cloud data received from a lidar module, of selecting a first to-be-combined image from among images received from a camera device that corresponds in time to the piece of point cloud data, selecting a second to-be-combined image from among the images that is the N.sup.th image before the first to-be-combined image in the time order, combining the first to-be-combined image and the second to-be-combined image to generate a combined image, generating a result image by incorporating the piece of point cloud data into the combined image, and inputting the result image into a trained machine learning model in order to determine a class to which each object in the result image belongs.

A data processing method, device and multi-sensor fusion method for multi-sensor fusion, which can group data captured by different sensors in different probe dimensions to simultaneous interpreting deep learning data based on pixel elements in the multi-dimensional matrix structure, thereby realize the more effective data mining and feature extraction to support more effective ability of environment perception and target detection.