公开(公告)号：EP4414746A1

公开(公告)日：2024-08-14

申请号：EP23198769.4

申请日：2023-09-21

申请人： Continental Autonomous Mobility Germany GmbH

发明人： Gheorghe, Ionut-Valentin ,  Kumari, Beena ,  Dr. Hellert, Christian

IPC分类号： G01S13/86 ,  G01S7/41 ,  G01S13/72 ,  G01S13/931 ,  G01S13/58 ,  G01S13/08 ,  G01S7/48

CPC分类号： G01S13/867 ,  G01S7/417 ,  G01S7/411 ,  G01S7/4808 ,  G01S13/08 ,  G01S13/58 ,  G01S13/726 ,  G01S13/931 ,  G06F18/251 ,  G06F18/253 ,  G06V10/806 ,  G06V10/803 ,  G06V10/82 ,  G06V20/58

摘要： The present invention is related to a multi-Object Detection and Tracking method and system based on AI techniques. According to invention, a computer-implemented method is provided for object detection and tracking of at least one object in an ego vehicle environment. The method uses a hybrid neural network adapted to:
(i) input raw sensor data, generated by at least one camera sensor and at least one radar sensor of the ego vehicle, namely raw camera data and raw radar data,
(ii) process independently raw camera data and raw radar data to generate object detections, to extract respective features from respective raw sensor data, meaning visual features extracted from raw camera data and depth features extracted from raw radar data, and to extract feature embeddings for each detected object,
(iii) fusion of the extracted visual and depth features, and generate fused object detections,
(iv) generating, from the extracted embeddings, tracks for each detected object, and
(v) associating the generated fused object detections to tracks.
In addition, the invention provides a computing system for end-to-end multi-object detection and tracking.