Abstract: Techniques are disclosed for improving classification performance in supervised learning. In accordance with some embodiments, a multiclass support vector machine (SVM) having three or more classes may be converted to a plurality of binary problems that then may be reduced via one or more reduced-set methods. The resultant reduced-set (RS) vectors may be combined together in one or more joint lists, along with the original support vectors (SVs) of the different binary classes. Each binary problem may be re-trained using the joint list(s) by applying a reduction factor (RF) parameter to reduce the total quantity of RS vectors. In re-training, different kernel methods can be combined, in accordance with some embodiments. Reduction may be performed until desired classification performance is achieved. The disclosed techniques can be used, for example, to improve classification speed, accuracy, class prioritization, or a combination thereof, in the SVM training phase, in accordance with some embodiments.

Abstract: A method of image processing is provided. The method may include: determining a candidate tuple from at least two images that are taken at different times, wherein the candidate tuples are determined using at least odometry sensor information. The couple of subsequent images have been detected by a moving image sensor moved by a vehicle. The odometry sensor information is detected by a sensor moved by the vehicle. The method may further include classifying the candidate tuples into a static tuple or a dynamic tuple. The static tuple represents a static object within the couple of subsequent images, and the dynamic tuple represents a moving object within the couple of subsequent images.

Abstract: An extrinsic camera calibration system (ECCS) comprising a surround view system (SVS) interface connecting the ECCS to the SVS. The SVS comprises four cameras mounted to front, back, left, and right sides of a vehicle respectively that are front, back, left, and right cameras, as well as an imaging interface at which images from the four cameras are output. A calibration pattern that comprises a set of features is used by an extrinsic calibration processor. A network interface is connected to the four cameras via the imaging interface. The processor receives, from each camera, an image comprising captured features from the calibration pattern. It determines and stores extrinsic calibration parameters (ECPs) for each camera that map coordinates of the features to camera coordinates and that are usable for subsequent normal operation of the cameras, the ECPs being determined from the image features.

Abstract: According to various examples, a vehicle controller is described comprising a determiner configured to determine information about surroundings of a vehicle, the information about the surroundings comprising information about velocities of objects in the surroundings of the vehicle and a velocity controller configured to input the information about the surroundings of the vehicle and a specification of a path of the vehicle to a convolutional neural network, to determine a target velocity of the vehicle along the path based on an output of the convolutional neural network and to control the vehicle according to the determined target velocity.

Abstract: Technologies for visualizing moving objects on a bowl-shaped image include a computing device to receive a first fisheye image generated by a first fisheye camera and capturing a first scene and a second fisheye image generated by a second fisheye camera and capturing a second scene overlapping with the first scene at an overlapping region. The computing device identifies a moving object in the overlapping region and modifies a projected overlapping image region to visualize the identified moving object on a virtual bowl-shaped projection surface. The projected overlapping image region is projected on the virtual bowl-shaped projection surface and corresponds with the overlapping region captured in the first and second fisheye images.

Abstract: According to various examples, a vehicle controller is described comprising a determiner configured to determine information about surroundings of a vehicle, the information about the surroundings comprising information about velocities of objects in the surroundings of the vehicle and a velocity controller configured to input the information about the surroundings of the vehicle and a specification of a path of the vehicle to a convolutional neural network, to determine a target velocity of the vehicle along the path based on an output of the convolutional neural network and to control the vehicle according to the determined target velocity.

Abstract: A perception device, including at least one image sensor configured to detect a plurality of images; an information estimator configured to estimate from each image of the plurality of images a depth estimate, a velocity estimate, an object classification estimate and an odometry estimate; a particle generator configured to generate a plurality of particles, wherein each particle of the plurality of particles comprises a position value determined from the depth estimate, a velocity value determined from the velocity estimate and a classification value determined from the classification estimate; an occupancy hypothesis determiner configured to determine an occupancy hypothesis of a predetermined region, wherein each particle of the plurality of particles contributes to the determination of the occupancy hypothesis.

Abstract: A method of determining a trajectory of motion of a vehicle in a predetermined region, wherein the predetermined region includes a plurality of sub-regions, the method being executed by one or more processors, the method including determining an occupancy hypothesis of the predetermined region, wherein the occupancy hypothesis indicates occupied sub-regions of the plurality of sub-regions and non-occupied sub-regions of the plurality of sub-regions; determining a utility value for each sub-region of the predetermined region; determining the trajectory of motion which crosses at least one sub-region of the non-occupied sub-regions, based on a function of the utility values of the least one sub-region of the non-occupied sub-regions crossed by the trajectory of motion and by maximizing a utility of motion of the vehicle, wherein the utility of motion of the vehicle is indicated by a function of the utility values of the sub-regions crossed by the trajectory of motion.

Abstract: A mechanism is described for facilitating depth and motion estimation in machine learning environments, according to one embodiment. A method of embodiments, as described herein, includes receiving a frame associated with a scene captured by one or more cameras of a computing device; processing the frame using a deep recurrent neural network architecture, wherein processing includes simultaneously predicating values associated with multiple loss functions corresponding to the frame; and estimating depth and motion based the predicted values.

Abstract: Technologies for determining a distance of an object from a vehicle include a computing device to identify an object captured in a fisheye image generated by a fisheye camera of the vehicle. The computing device projects a contour of the identified object on a selected virtual plane that is located outside the vehicle and selected from a predefined set of virtual planes based on a location of the identified object relative to the vehicle. The computing device identifies a bottom of the projected contour on the selected virtual plane and determines an intersection point of an imaginary line with a ground plane coincident with a plane on which the vehicle is positioned. The imaginary line passes through each of the identified bottom of the projected contour and the fisheye camera. The computing device determines a location of the identified object relative to the vehicle based on the determined intersection point and the identified bottom of the projected contour.

Abstract: An extrinsic camera calibration system (ECCS) comprising a surround view system (SVS) interface connecting the ECCS to the SVS. The SVS comprises four cameras mounted to front, back, left, and right sides of a vehicle respectively that are front, back, left, and right cameras, as well as an imaging interface at which images from the four cameras are output. A calibration pattern that comprises a set of features is used by an extrinsic calibration processor. A network interface is connected to the four cameras via the imaging interface. The processor receives, from each camera, an image comprising captured features from the calibration pattern. It determines and stores extrinsic calibration parameters (ECPs) for each camera that map coordinates of the features to camera coordinates and that are usable for subsequent normal operation of the cameras, the ECPs being determined from the image features.

Abstract: A method for flexibly sintering rare earth permanent magnetic alloy comprises: (1) weighing fine powder of rare earth permanent magnetic alloy, loading the fine powder in molds, and orientedly compacting the fine powder in a press machine and in inert atmosphere to obtain blanks and loading the blanks into charging boxes; (2) after air between the second conveying vehicle and the first isolating valve of the glove box is replaced with inert gas, opening the two isolating valves connected with each other; wherein after a first rolling wheel transmission in the second conveying vehicle transfers the charging tray into the first chamber of the glove box, the two isolating valves are closed, and the second conveying vehicle leaves; (3) after a first conveying vehicle is coupled with a third isolating valve at an end of the second chamber, locking two matching flanges of the two isolating valves tightly; (4) after the first conveying vehicle is coupled with an isolating valve of a sintering furnace, locking matchi

Abstract: A perception device including a receiver configured to receive sensor information including information about a location of one or more objects detected by a sensor; a memory configured to store an occupancy grid of a predetermined region, wherein the occupancy grid includes a plurality of grid cells, wherein each grid cell represents an area in the predetermined region, wherein at least one grid cell of the plurality of grid cells is associated with a respective single occupancy hypothesis; a single occupancy hypothesis determiner configured to determine a single occupancy hypothesis; wherein the single occupancy hypothesis includes degree of belief of occupancy of the grid cell depending on the sensor information; wherein a contribution of a sensor information value to the degree of belief of occupancy substantially decreases with an increase of a distance of the location of the object detected by the sensor from a center of the grid cell.

Abstract: A method of image processing is provided. The method may include: determining a candidate tuple from at least two images that are taken at different times, wherein the candidate tuples are determined using at least odometry sensor information. The couple of subsequent images have been detected by a moving image sensor moved by a vehicle. The odometry sensor information is detected by a sensor moved by the vehicle. The method may further include classifying the candidate tuples into a static tuple or a dynamic tuple. The static tuple represents a static object within the couple of subsequent images, and the dynamic tuple represents a moving object within the couple of subsequent images.

Abstract: A perception device, including at least one image sensor configured to detect a plurality of images; an information estimator configured to estimate from each image of the plurality of images a depth estimate, a velocity estimate, an object classification estimate and an odometry estimate; a particle generator configured to generate a plurality of particles, wherein each particle of the plurality of particles comprises a position value determined from the depth estimate, a velocity value determined from the velocity estimate and a classification value determined from the classification estimate; an occupancy hypothesis determiner configured to determine an occupancy hypothesis of a predetermined region, wherein each particle of the plurality of particles contributes to the determination of the occupancy hypothesis.

Abstract: A method for sensing the position of a vehicle on a three-dimensional map is provided, which may include receiving, from one or more sensors, sensor information about a three-dimensional position of the vehicle; and applying a distributed Kalman filter to the sensor information to determine the position of the vehicle on the three-dimensional map.

Abstract: An occupancy grid object determining device is provided, which may include a grid generator configured to generate an occupancy grid of a predetermined region, the occupancy grid including a plurality of grid cells and at least some of the grid cells having been assigned an information about the occupancy of the region represented by the respective grid cell, a determiner configured to determine at least one object in the occupancy grid wherein the at least one object includes a plurality of grid cells, and a remover configured to remove occupancy information from at least one grid cell of the plurality of grid cells of the determined object.

Abstract: A visual odometry device, including: an image sensor configured to provide a first image and a second image; a visual feature extractor configured to extract at least three visual features corresponding to each of the first image and the second image; and a position determiner, configured to determine a change of a position of the at least three visual features between the first image and the second image, and to determine a degree of translation of the visual odometry device based on the determined change of position.

Abstract: Various aspects of this disclosure provide an area occupancy determining device. The device may include a memory configured to store at least one occupancy grid of a predetermined region, and a processor. The processor may be configured to generate the occupancy grid of the predetermined region. The occupancy grid includes a plurality of grid cells, each grid cell framed by respective grid cell frame lines. At least some of the grid cells have been assigned an information about the occupancy of the region represented by the respective grid cell. The processor may further be configured to dynamically update the occupancy grid, thereby successively generating a plurality of updated occupancy grids. Each updated occupancy grid is moved relative to the previous occupancy grid such that an origin coordinate of the updated occupancy grid is positioned on a contact point of grid cell frame lines of adjacent grid cells.

Abstract: System and techniques for a spiking neural network and support vector machine hybrid classifier are described herein. A first set of sensor data may be obtained, e.g., from a corpus of sample sensor data. A feature set is extracted from the sensor data using a spiking neural network (SNN). A support vector machine (SVM) may then be created for the sensor data using the feature set. The SVM may then be used to classify a second set of sensor data.