Method and system for object recognition based on a trainable dynamic system

• US 8,705,849 B2
• Filed: 11/24/2008
• Issued: 04/22/2014
• Est. Priority Date: 11/24/2008
• Status: Active Grant

• Current Assignee: Toyota Jidosha Kabushiki Kaisha (Toyota Motor Corporation)
• Original Assignee: Toyota Motor Engineering & Manufacturing North America Incorporated (Toyota Motor Corporation)
• Inventors: Prokhorov, Danil V.
• Primary Examiner(s): BHATNAGAR, ANAND P
• Application Number:
  US12/276,703
• Publication Number:
  US 20100128975A1
• Time in Patent Office: 1,975 Days
• Field of Search: 382/100, 382/103, 382/104, 382/105, 382/106, 382/107, 382155-159, 382/181, 382/190, 382/195, 382/201, 348/135, 348/143, 348169-172
• US Class Current: 382/155
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 20/10 Terrestrial scenes scenes under surveillance with static cameras G06V20/52; scenes perceived from the exterior of a vehicle G06V20/56; scenes perceived from the interior of a vehicle G06V20/59
• CPC Class Codes: G06V 20/64 Three-dimensional objects

Abstract

A system for object recognition in which a multi-dimensional scanner generates a temporal sequence of multi-dimensional output data of a scanned object. That data is then coupled as an input signal to a trainable dynamic system. The system exemplified by a general-purpose recurrent neural network is previously trained to generate an output signal representative of the class of the object in response to a temporal sequence of multi-dimensional data.

Claims

1. A system for object recognition comprising:

a multi-dimension distance scanner which generates a temporal sequence of multi-dimensional output data of a scanned object,

a trainable recurrent neural network which receives the temporal sequence of multi-dimensional output data as input data, said trainable recurrent neural network being trained to generate an output signal representative of a class of the scanned object,

wherein said trainable recurrent neural network is trained for rotation invariant recognition of the scanned object,

wherein the trainable recurrent neural network is trained for rotation invariant recognition in preset rotational increments,

wherein the trainable recurrent neural network is trained for rotation invariant recognition simultaneously in all preset rotational increments.

2. The invention as defined in claim 1 and comprising means for normalizing said multi-dimensional output data.

3. A method for object recognition comprising the steps of:

creating a temporal sequence of multi-dimensional output data of a scanned object,

coupling said multi-dimensiional output data as input data to a trainable recurrent neural network previously trained to identify different classes of objects,

wherein said trainable recurrent neural network is trained for rotation invariance,

wherein said trainable recurrent neural network is trained for rotation invariance in preset rotational steps,

wherein said trainable recurrent neural network is trained for rotation invariance simultaneously in all rotational steps.