A signal processing apparatus including a first position calculation unit that calculates a three-dimensional position of a target on a first coordinate system from a stereo image captured by a stereo camera, a second position calculation unit that calculates a three-dimensional position of the target on a second coordinate system from a sensor signal of a sensor capable of obtaining position information of at least one of a lateral direction and a longitudinal direction and position information of a depth direction, a correspondence detection unit that detects a correspondence relationship between the target on the first coordinate system and the target on the second coordinate system, and a positional relationship information estimating unit that estimates positional relationship information of the first coordinate system and the second coordinate system on the basis of the detected correspondence relationship.

A method is described for identifying and classifying objects, detected by a sensor apparatus which actively emits radiation, in terms of the relevance thereof to a driving situation of a moving vehicle, wherein radiation is emitted by the sensor apparatus and the echo radiation reflected at objects is received as measurement values, including: detecting measurement values in relation to the driving situation of the vehicle, performing an analysis of the driving situation represented by the measurement values and identifying at least one possible object, classifying the at least one identified object in an object class of a plurality of object classes, in which performing the analysis of the measurement values includes: transforming the detected measurement values from a coordinate system fixed in terms of the vehicle into a coordinate system fixed in terms of space for generating measurement values fixed in terms of spatial coordinates, wherein this transformation is based on the vehicle speed and the yaw rate of the vehicle in the determined driving situation, subdividing at least one total area, which is situated in the detection region of the sensor apparatus emitting radiation and which is coplanar or parallel with the roadway surface, into a plurality of partial areas, wherein partial areas adjoining one another partly overlap, determining the number and/or the statistical dispersion of the detected measurement values fixed in terms of spatial coordinates for each one of these partial areas, and in which performing the analysis of the driving situation represented by the measurement values and identifying at least one possible object includes: comparing the number and/or and statistical dispersion of the detected measurement values fixed in terms of spatial coordinates for each one of these partial areas, in each case with characteristic patterns, and identifying at least one object possibly present in a partial area and classifying the at least one identified object in an object class of a plurality of object classes depending on this comparison.

Systems, methods and computer-readable media are provided for tracking a moving sports object. In one example, a method of tracking a moving sports object includes calibrating a perspective of an image of a camera to a perspective of a Doppler radar for simultaneous tracking of the moving sports object, and tracking the moving sports object simultaneously with the camera and Doppler radar. The method may further comprise removing offsets or minimizing differences between simultaneous camera measurements and Doppler radar measurements of the moving sports object. The method may also include combining a camera measurement of an angular position of the moving sports object with a simultaneous Doppler measurement of a radial distance, speed or other measurement of the moving sports object.

Techniques and apparatuses are described that implement a smart-device-based radar system capable of detecting a frame-of-reference change. In particular, a radar system includes a frame-of-reference machine-learned module trained to recognize whether or not the radar system's frame of reference changes. The frame-of-reference machine-learned module analyzes complex radar data generated from at least one chirp of a reflected radar signal to analyze a relative motion of at least one object over time. By analyzing the complex radar data directly using machine learning, the radar system can operate as a motion sensor without relying on non-radar-based sensors, such as gyroscopes, inertial sensors, or accelerometers. With knowledge of whether the frame-of-reference is stationary or moving, the radar system can determine whether or not a gesture is likely to occur and, in some cases, compensate for the relative motion of the radar system itself.

A computer implemented method for determining the position of a vehicle, wherein the method comprises: determining at least one scan comprising a plurality of detection points, wherein each detection point is evaluated from a signal received at the at least one sensor and representing a location in the vehicle environment; determining, from a database, a predefined map, wherein the map comprises a plurality of elements in a map environment, each of the elements representing a respective one of a plurality of static landmarks in the vehicle environment, and the map environment representing the vehicle environment; matching the plurality of detection points and the plurality of elements of the map; determining the position of the vehicle based on the matching; wherein the predefined map further comprises a spatial assignment of a plurality of parts of the map environment to the plurality of elements, and wherein the spatial assignment is used for the matching.

In various examples, a deep neural network(s) (e.g., a convolutional neural network) may be trained to detect moving and stationary obstacles from RADAR data of a three dimensional (3D) space, in both highway and urban scenarios. RADAR detections may be accumulated, ego-motion-compensated, orthographically projected, and fed into a neural network(s). The neural network(s) may include a common trunk with a feature extractor and several heads that predict different outputs such as a class confidence head that predicts a confidence map and an instance regression head that predicts object instance data for detected objects. The outputs may be decoded, filtered, and/or clustered to form bounding shapes identifying the location, size, and/or orientation of detected object instances. The detected object instances may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

A processor in an object detection device is configured to: acquire settings data for one or more radar grids, each radar grid being set for a measurement area of a corresponding radar and consisting of radar cells; acquire settings data for a processing grid for clustering operations, the processing grid being set for the one or more measurement areas and consisting of processing cells; calculate, for each processing cell, one or more likelihoods associated with measurements of one or more related radar cells based on distances between the processing cell and the one or more related radar cells; calculate a grid value of each of the processing cells in the processing grid based on the one or more likelihoods; and perform a clustering operation on each processing cell based on distances between and/or grid values of the processing cell and one or more different processing cells.

A method (200) of determining a transformation matrix for transformation of ranging data from a first coordinate system for the ranging sensor to a second coordinate system for an image sensor is disclosed. The method comprises providing (201) a ranging sensor and an image sensor; acquiring (202) a ranging frame sequence, and an image frame sequence; determining (203) points of motion in frames of each acquired frame sequence; for each frame in one of the frame sequences: evaluating (204) if a single motion point has been determined in the frame, and if a single motion point has been determined, evaluating (206) if a single motion point has been determined in a temporally corresponding frame of the other frame sequence and, in that case, pairing (207) the temporally corresponding frames, whereby a set of frame pairs is formed, and determining (209) the transformation matrix based on the set of frame pairs.

A computer-implemented method for determining a position of a vehicle is disclosed, wherein the vehicle is equipped with a sensor for capturing scans of a vicinity of the vehicle, wherein the method comprises at least the following steps carried out by computer-hardware components: capturing at least one scan by means of the sensor with a plurality of sensor data samples given in a sensor data representation; determining, from a database, a predefined map with at least one element is given in a map data representation; determining a transformed map by transforming the at least one element of the predefined map from the map data representation into the sensor data representation; matching at least a subset of the sensor data samples of the at least one scan and the at least one element of the transformed map; and determining the position of the vehicle based on the matching.

An evaluation method for radar measurement data of a mobile radar measurement system includes the steps of preparing a multidimensional range-Doppler map from the radar measurement data. In this evaluation method, each multidimensional range-Doppler map is stored together with time information. Moreover, at least one multidimensional range-Doppler map with time information is propagated on the basis of known movement data of the radar measurement system to the current time. The multiple multidimensional range-Doppler maps may be combined to form a combined range-Doppler map.