Single component data processing system and method utilizing a trained neural network

Patent metadata

Field	Value
Publication number	US-11853812-B2
Application number	US-201816228024-A
Country	US
Kind code	B2
Filing date	Dec 20, 2018
Priority date	Dec 20, 2018
Publication date	Dec 26, 2023
Grant date	Dec 26, 2023

Abstract

Methods, systems, apparatuses, and computer program products are provided that are configured to processor sensor data using a single component specialized sensor data processing neural network. The single component data processing system is configured to receive input data from a sensor, analyze the input data using a neural network embodied by a single trained sensor data processing component, wherein the single trained sensor data processing component is trained to product output data that approximates a plurality of task-specific transformations performed in a pipeline manner, and produce the output data following transformation of the input data. Additionally, methods, systems, apparatuses, and computer program products are provided for training a single trained sensor data processing component to approximate a plurality of task-specific transformations performed in a pipeline manner.

First claim

What is claimed is:

1. A single component sensor data processing system configured to:
receive input data from a particular sensor of a plurality of sensors associated with a vehicle;
analyze the input data using a neural network embodied on a particular single integrated chip corresponding to the particular sensor from which the input data was received, wherein the neural network on the single integrated chip is trained to produce output data that approximates a plurality of task-specific transformations performed in a serial pipeline, the plurality of task-specific transformations corresponding to the particular sensor to transform the input data specific to the particular sensor into the output data that is interpretable by a particular second system that processes data from each of the plurality of sensors, wherein serial pipeline recieves training input data corresponding to the input data as input and generates processed data as output, and wherein the neural network is trained based on the training input data and the processed data; and
produce the output data following the transformations of the input data.

2. The single component sensor data processing system according to claim 1 , wherein the single integrated chip is further configured to output the output data to the particular second system.

3. The single component sensor data processing system according to claim 1 , wherein the plurality of task-specific transformations comprises a plurality of sequentially executed transformations corresponding to a sensor type of the particular sensor.

4. The single component sensor data processing system according to claim 1 , wherein the input data comprises raw data.

5. The single component sensor data processing system according to claim 1 , wherein the processed dataset is generated from the input dataset by processing each training input data of the input dataset using a separate hardware component for each task-specific transformation of the plurality of task-specific transformations performed in the serial pipeline.

6. The single component sensor data processing system according to claim 1 , wherein the plurality of task-specific transformations performed in the serial pipeline comprises an image correction transformation, a noise reduction transformation, an image scaling transformation, a gamma correction transformation, an image enhancement transformation, a color-space conversion transformation, a chroma subsampling transformation, a framerate conversion transformation, and an image compression transformation.

7. The single component sensor processing system according to claim 1 , wherein the sensor data is received directly from the particular sensor.

8. The single component sensor processing system according to claim 1 , wherein the plurality of task-specific transformations comprises an additional transformation that highlights a region of interest within the input data.

9. The single component sensor processing system according to claim 1 , wherein the particular second system comprises a perception system configured to perform a field environment analysis based at least in part on a plurality of output data produced via processing of a plurality of portions of input data from the plurality of sensors.

10. A method for processing data comprising:
configuring a neural network to perform a plurality of task-specific transformations;
receiving input data from a particular sensor of a plurality of sensors associated with a vehicle;
analyzing the input data using the neural network embodied on a particular single integrated chip corresponding to the particular sensor from which the input data was received, wherein the neural network on the single integrated chip is trained to produce output data that approximates the plurality of task-specific transformations performed in a serial pipeline, the plurality of task-specific transformations corresponding to the particular sensor to transform the input data specific to the particular sensor into the output data that is interpretable by a particular second system that processes data from each of the plurality of sensors, wherein serial pipeline recieves training input data corresponding to the input data as input and generates processed data as output, and wherein the neural network is trained based on the training input data and the processed data; and
producing, from the single integrated chip, the output data following the transformations of the input data utilizing the single integrated chip.

11. The method for processing data according to claim 10 , further comprising outputting, via the single integrated chip, the output data to the particular second system.

12. The method for processing data according to claim 10 , wherein the plurality of task-specific transformations comprises a plurality of sequentially executed transformations corresponding to a sensor type of the particular sensor.

13. A method for training a neural network embodied on a single integrated chip, the method comprising:
receiving training input data from an input dataset collected from at least one sensor of a plurality of sensors associated with a vehicle;
receiving processed data from a processed dataset created by a sensor data processing pipeline system comprising a plurality of task-specific transformation components for performing a plurality of task-specific transformations in a serial pipeline, wherein the serial pipeline recieves training input data as input and generates the processed data as output; and
training the neural network embodied on the single integrated chip to approximate the plurality of task-specific transformations from the training input data to the processed data, wherein the neural network is trained based on the training input data and the processed data, and wherein the single integrated chip corresponds to the at least one sensor, and wherein the plurality of task-specific transformations correspond to the at least one sensor to transform the training input data specific to the particular sensor into the processed data that are interpretable by a particular second system that processes data from each of the plurality of sensors.

14. The method for training a neural network according to claim 13 , wherein the training input dataset is a pre-collected input value dataset.

15. The method for training a neural network according to claim 13 , wherein the training input dataset is a dataset collected in real-time.

16. The method for training a neural network according to claim 13 , wherein the processed dataset is a pre-collected dataset.

17. The method for training a neural network according to claim 13 , wherein the training input data comprises raw data.

18. The method for training a neural network according to claim 13 , wherein the training input data comprises data that has undergone at least one pre-processing transformation.

Assignees

• Here Global Bv

Inventors

• Viswanathan Anirudh

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06F9/54 Primary

  Interprogram communication · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/08

  Learning methods · CPC title