A detection system includes an acquisition portion scanning light to acquire point-cloud information corresponding to a plurality of positions of a detection target object; an estimation portion using consistency with an outer shape model of the detection target object to estimate a location and attitude of the detection target object based on the point-cloud information; and an output portion outputting information relating to a movement target location based on an estimation result, wherein the estimation portion fits an outer shape model indicating an outer shape of the detection target object to a point cloud according to the point-cloud information, and uses point-cloud information existing outside the outer shape model to estimate the location and the attitude of the detection target object.

A LIDAR-to-LIDAR alignment system includes a memory and an autonomous driving module. The memory stores first and second points based on outputs of first and second LIDAR sensors. The autonomous driving module performs a validation process to determine whether alignment of the LIDAR sensors satisfy an alignment condition. The validation process includes: aggregating the first and second points in a vehicle coordinate system to provide aggregated LIDAR points; based on the aggregated LIDAR points, performing (i) a first method including determining pitch and roll differences between the first and second LIDAR sensors, (ii) a second method including determining a yaw difference between the first and second LIDAR sensors, or (iii) point cloud registration to determine rotation and translation differences between the first and second LIDAR sensors; and based on results of the first method, the second method or the point cloud registration, determining whether the alignment condition is satisfied.

A method of classifying an object according to an embodiment includes extracting a first feature by transforming rectangular coordinates of points included in the box of the object, obtained from a point cloud acquired using a LiDAR sensor, into complex coordinates and performing Fast Fourier Transform (FFT) on the complex coordinates, obtaining an average and a standard deviation as a second feature, the average and the standard deviation being parameters of a Gaussian model for the points included in the box of the object, and classifying the type of object based on at least one of the first feature or the second feature.

A method of tracking an object using a LiDAR sensor includes: acquiring a point cloud using each of LiDAR sensors and clustering the point cloud. The clustering includes generating a three-dimensional (3D) grid map using the point cloud acquired by each of the LiDAR sensors; labeling voxels present in the 3D grid map with one cluster using direction points of the voxels; and checking whether it is necessary to label the voxels with mutually different clusters.

A system for identifying a location of one or more assets in a predefined two-dimensional area comprises at least three tracking stations and one or more tracking tags. Each tracking station selectively emits a vertical laser line upon which is embedded a unique identifier, selectively sweeps its laser line about its central axis such that each tracking station's laser line sweeps across at least a portion of the predefined 2-D area, and selectively transmits a current angle of its laser line as its laser line sweeps about its central axis. Each tracking tag detects a laser line from at least three tracking stations within its line of sight. Each tracking tag decodes the unique tracking station identifier, receives the current angle from the tracking station corresponding to the detected laser line, and stores the decoded unique tracking station identifier and the received current angle.

In one embodiment, a method for solution inference using neural networks in LiDAR localization includes constructing a cost volume in a solution space for a predicted pose of an autonomous driving vehicle (ADV), the cost volume including a number of sub volumes, each sub volume representing a matching cost between a keypoint from an online point cloud and a corresponding keypoint on a pre-built point cloud map. The method further includes regularizing the cost volume using convention neural networks (CNNs) to refine the matching costs; and inferring, from the regularized cost volume, an optimal offset of the predicted pose. The optimal offset can be used to determine a location of the ADV.

A LIDAR-based method of determining an absolute speed of an object at a relatively longer distance from an ego vehicle, including: estimating a self speed of the ego vehicle using a first frame t-1 and a second frame t obtained from a LIDAR sensor by estimating an intervening rotation θ about a z axis and translation in orthogonal x and y directions using a deep learning algorithm over a relatively closer distance range; dividing each of the first frame t-1 and the second frame t into multiple adjacent input ranges and estimating a relative speed of the object at the relatively longer distance by subsequently processing each frame using a network, with each input range processed using a corresponding convolutional neural network; and combining the estimation of the estimating the self speed with the estimation of the estimating the relative speed to obtain an estimation of the absolute speed.

A process for sensing a scene. The process includes receiving sensor data from a plurality of sensor modalities, where each sensor modality observes at least a portion of the scene containing at least one of the objects of interest and generates sensor data conveying information on the scene and of the object of interest. The process further includes processing the sensor data from each sensor modality to detect objects of interest and produce a plurality of primary detection results, each detection result being associated with a respective sensor modality. The process also includes fusing sensor data from a first sensor modality with sensor data from a second sensor modality to generate a fused 3D map of the scene, processing the fused 3D map to detect objects of interest and produce secondary detection results and performing object level fusion on the primary and the secondary detection results.

A measurement vehicle (101) acquires measurement environment data indicating a measurement environment from a measurement system and transmits the acquired measurement environment data to a base station (104). The measurement vehicle (101) receives movement measurement instruction data indicating an instruction on the movement measurement from the base station (104). The measurement vehicle (101) controls the measurement system in accordance with the instruction indicated by the received movement measurement instruction data.

The disclosure relates to a method and system for determining a moving/stationary state of a tracking target based on a neural network. The method includes: a training step of extracting feature points for each tracking target in a tracking target set, converting the feature points into feature vectors and labeling the feature vectors with truth values of moving and stationary states of the tracking target, and training a neural network by using the feature vectors and the corresponding truth values of the moving and stationary states as training sample parameters, to obtain a trained neural network classification model; and a testing step of extracting feature vectors of the tracking target, and then inputting the feature vectors into the neural network classification model, to obtain a moving and stationary state classification result for the tracking target. According to the disclosure, the use of the neural network method for determination of a moving/stationary state of the tracking target from the laser radar can improve the accuracy of a determination result.