METHOD FOR TRACKING OBJECTS AT THE RECOGNITION STAGE FOR UNMANNED VEHICLES Russian patent published in 2023 - IPC G01S17/93 

FIELD: tracking.

SUBSTANCE: claimed invention relates to computer-implemented methods and systems for determining the presence of an object, namely, methods for tracking objects in the environment of unmanned vehicles. The method for tracking objects at the recognition stage for unmanned vehicles includes combining the output data of various recognition systems and combining them into a single output, in which objects are recognized by image recognition systems and lidar clouds: image recognition systems take a raw image as input and calculate the probabilities of classes and bounding boxes for various objects in the image, the network uses weights that are pre-trained on the dataset using a neural network, a list of recognitions is obtained, which includes: a bounding box, classification of the recognized object and confidence in the forecast obtained, recognition by lidar clouds is performed by input signals raw lidar point clouds, in which the neural network takes a lidar point cloud as input and performs transformations to find the central points and bounding boxes of objects. Camera detections are also post-processed, including the conversion of objects recognized by the camera, represented as 2D bounding boxes, into 3D objects, for use in association at the tracking stage: lidar data (a point cloud) is used to perform the transformation. To simplify the computational complexity, it undergoes preprocessing, consisting of the following steps: filtering the point cloud based on the camera's field of view; voxylesation of the filtered point cloud; segmentation of points belonging to the ground; to accurately perform target tracking, an association of recognitions obtained using heterogeneous sensors is performed, including a dynamic search for the shortest path along the residual graph.

EFFECT: increased performance of object tracking at the recognition stage for unmanned vehicles.

4 cl, 10 dwg, 3 tbl