Method for common detecting, trackng and classifying of objects
AssigneeOSRAM GmbH
InventorsSikandar AMIN, Bharti MUNJAL, Meltem Demirkus BRANDLMAIER, Abdul Rafey AFTAB, Fabio GALASSO
Life of this patent
Estimating the term…
Abstract
A method for machine-based training of a computer-implemented network for common detecting, tracking, and classifying of at least one object in a video image sequence having a plurality of successive individual images. A combined error may be determined during the training, which error results from the errors of the determining of the class identification vector, determining of the at least one identification vector, the determining of the specific bounding box regression, and the determining of the inter-frame regression.
Claims21
A method for machine-based training of a computer-implemented network for common detecting, tracking, and classifying of at least one object in a video image sequence having a multiplicity of successive frames, wherein the method comprises:
- providing a multiplicity frame pairs, wherein each frame pair comprises a first and a succeeding second frame, wherein each frame has at least one object proposal, to which a respective class identification vector is uniquely assigned;
- for each object proposal:
- determining a feature vector assigned to the class identification vector;
- determining an identification vector;
- determining a class identification vector-specific bounding box regression;
- determining a correlated feature map between the first frame and the succeeding second frame;
- determining an inter-frame regression for each object proposal;
- minimizing a combined error resulting from the errors in determining the class identification vector, determining the identification vector, determining the specific bounding box regression and determining the inter-frame regression.
A method for common detecting, tracking, and classifying of at least one object in a video image sequence having a multiplicity of successive frames by means of a trained computer-implemented network;
- wherein the method comprises:
- receiving a first frame and a succeeding second frame;
- detecting at least one object in the first frame and the succeeding second frame;
- selecting an object from the first and second frames;
- ascertaining at least one classification vector and a position for the object from the first and second frames;
- ascertaining an association value on the basis of the ascertained classification vector and the position;
- generating a temporarily consistent and unique identification vector of the at least one object for each frame in response to the ascertained association value;
- wherein a relative weighting between the ascertained classification vector and the position, said weighting being dependent on the time between the first and second frames, is provided for the generation of the association value.
An arrangement having:
- a buffer memory for at least one first and one second frame of a video sequence;
- an RPN module for generating a first multiplicity of independent object proposals from the first frame and a second multiplicity of independent object proposals from the second frame;
- a determination and identification module trained jointly with a prediction module, said determination and identification module being coupled to the RPN module and being configured to identify objects from the first multiplicity and the second multiplicity of object proposals and to detect them in the first and second frames, and being further configured to make a prediction of the position of an object in the first frame and in the second frame from the object proposals;
- the prediction module trained jointly with the determination and identification module, said prediction module being coupled to the RPN module and being configured to make an inter-frame prediction for each object in the first frame on the basis of the first multiplicity of object proposals and the second multiplicity of object proposals.
21-26. (canceled)
Dependencies are inferred from the claim text.