IPC분류정보
국가/구분 |
United States(US) Patent
등록
|
국제특허분류(IPC7판) |
|
출원번호 |
US-0276075
(2008-11-21)
|
등록번호 |
US-8175782
(2012-05-08)
|
우선권정보 |
EP-07121776 (2007-11-28); EP-08161696 (2008-08-04) |
발명자
/ 주소 |
- Gepperth, Alexander
- Fritsch, Jan Nikolaus
|
출원인 / 주소 |
- Honda Research Institute Europe GmbH
|
대리인 / 주소 |
|
인용정보 |
피인용 횟수 :
2 인용 특허 :
0 |
초록
▼
A computer-implemented system and method for estimating properties of objects represented in digital images, comprising the steps of (a) encoding input data from a sensor in a neural map comprising neurons having numerical activation values, wherein the activation values in the neural maps have cont
A computer-implemented system and method for estimating properties of objects represented in digital images, comprising the steps of (a) encoding input data from a sensor in a neural map comprising neurons having numerical activation values, wherein the activation values in the neural maps have continuous time dynamics defined by an update scheme; (b) creating, adapting and deleting weights of the neural map in unsupervised, incremental manner; (c) transmitting data from an input map to an output map, based on the values of the weights; wherein each weight between the input map (IM) and a neural output map (OM) has a unique source and destination neuron; and wherein data transmission is directed; and (d) detecting correlations between the input map (IM).
대표청구항
▼
1. A computer-implemented method for a driver assistance system for estimating properties of objects in a vehicle's environment represented in digital images received from sensors, the properties including at least one of object identity, object position, and object distance, comprising the steps of
1. A computer-implemented method for a driver assistance system for estimating properties of objects in a vehicle's environment represented in digital images received from sensors, the properties including at least one of object identity, object position, and object distance, comprising the steps of: in a neural map comprising neurons having numerical activation values, extracting data on object identity and data on one of the other properties of the object from the input data from the sensors, and assigning the data on the object identity to a first neural map and assigning the data on said one of the other properties to a second neural map;the data on the object identity being encoded in accordance with common data encoding scheme (CDES) wherein two-dimensional arrays of numbers are encoded into a set of two-dimensional arrays of floating point numbers;between the first and the second neural maps, performing an unsupervised learning for adapting weights that implement a unique connection between a source neuron and a destination neuron, and incrementally increasing the complexity of a learning task;transmitting data between the first and the second neural maps, based on the values of the weights;detecting correlations between the first and the second neural maps, and estimating object identity from the second neural map based on the correlations; anddetermining the object identity based on the first neural map and said estimation of the object identity. 2. The method of claim 1, wherein the step of extracting is realized by a population coding scheme. 3. A driver assistance system for vehicles or planes having computing means designed for implementing a method for a driver assistance system for estimating properties of objects in a vehicle's or plane's environment represented in digital images received from sensors, the properties including at least one of object identity, object position, and object distance, comprising the steps of: in a neural map comprising neurons having numerical activation values, extracting data on object identity and data on one of the other properties of the object from the input data from the sensors, and assigning the data on the object identity to a first neural map and assigning the data on said one of the other properties to a second neural map;the data on the object identity being encoded in accordance with common data encoding scheme (CDES) wherein two-dimensional arrays of numbers are encoded into a set of two-dimensional arrays of floating point numbers;between the first and the second neural maps, performing an unsupervised learning for adapting weights that implement a unique connection between a source neuron and a destination neuron, and incrementally increasing the complexity of a learning task;transmitting data between the first and the second neural maps, based on the values of the weights;detecting correlations between the first and the second neural maps, and estimating object identity from the second neural map based on the correlations; anddetermining the object identity based on the first neural map and said estimation of the object identity. 4. A non-transitory computer readable medium comprising computer executable code which when executed by a computer performs a method for a driver assistance system for estimating properties of objects in a vehicle's environment represented in digital images received from sensors, the properties including at least one of object identity, object position, and object distance, comprising the steps of: in a neural map comprising neurons having numerical activation values, extracting data on object identity and data on one of the other properties of the object from the input data from the sensors, and assigning the data on the object identity to a first neural map and assigning the data on said one of the other properties to a second neural map;the data on the object identity being encoded in accordance with common data encoding scheme (CDES) wherein two-dimensional arrays of numbers are encoded into a set of two-dimensional arrays of floating point numbers;between the first and the second neural maps, performing an unsupervised learning for adapting weights that implement a unique connection between a source neuron and a destination neuron, and incrementally increasing the complexity of a learning task;transmitting data between the first and the second neural maps, based on the values of the weights;detecting correlations between the first and the second neural maps, and estimating object identity from the second neural map based on the correlations; anddetermining the object identity based on the first neural map and said estimation of the object identity. 5. A device for estimating properties of objects in a vehicle's environment represented in digital images received from sensors, comprising: extracting means, in a neural map comprising neurons having numerical activation values, for extracting data on object identity and data on one of the other properties of the object from the input data from the sensors, and assigning the data on the object identity to a first neural map and assigning the data on said one of the other properties to a second neural map;encoding means for encoding the data on the object identity in accordance with common data encoding scheme (CDES) wherein two-dimensional arrays of numbers are encoded into a set of two-dimensional arrays of floating point numbers;unsupervised learning means for performing an unsupervised learning for adapting weights, to the first and the second neural maps, that implement a unique connection between a source neuron and a destination neuron, and incrementally increasing the complexity of a learning task;transmitting means for transmitting data between the first and the second neural maps, based on the values of the weights;detection means for detecting correlations between the first and the second neural maps, and estimating object identity from the second neural map based on the correlations; anddetermining means for determining the object identity based on the first neural map and said estimation of the object identity. 6. The device of claim 5, for use in a driver assistance system in a vehicle. 7. The device of claim 6, for controlling emergency braking maneuvers of a vehicle.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.