Recurrent Neural Network for Vehicle Object Position Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current advanced driver assistance systems and self-driving technologies face limitations in predicting the movement of objects, as they rely on explicit features like speed and trajectory, which restricts their ability to accurately anticipate future positions of pedestrians and other vehicles, potentially leading to accidents.

Innovation Solution

The method employs a recurrent neural network, specifically a Long Short-Term Memory (LSTM) network, to capture and process current and past data frames from automotive imaging sensors, predicting the future position of moving objects within the vehicle's scenery without the need for explicit feature extraction, using probability maps to identify potential future locations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If standard approaches use explicit feature extraction (speed, trajectory) for prediction, then the prediction process is simple and interpretable, but the prediction accuracy is limited

Engineering Contradiction:
Improveprediction accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical feature extraction and prediction methods with a neural network-based system. Instead of manually extracting speed and trajectory features and applying prediction algorithms, the system uses a neural network to automatically learn and extract relevant features from raw sensor data, substituting the mechanical approach with an intelligent, adaptive system that achieves higher prediction accuracy.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The neural network performs self-service by automatically extracting features and making predictions without requiring manual feature engineering. The system learns the optimal features and prediction patterns from training data, enabling it to serve its own feature extraction needs and improve accuracy without additional human intervention or complex preprocessing pipelines.

Inventive Principle:
Principle #25Self-service

2Measurement precision

If more sensor data is processed to improve prediction accuracy, then the prediction becomes more accurate, but the computational load increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational energy consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary action by pre-training the neural network offline with large datasets. This preliminary training phase allows the network to learn optimal feature representations and prediction patterns in advance, so that during actual operation, the system can make accurate predictions with minimal real-time computational overhead, reducing energy consumption during critical prediction moments.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3278317B1Method and electronic device
Publication Date: 2022.11.16 SONY GROUP CORP
  • EP3278317B1 patent drawingFigure 1
  • EP3278317B1 patent drawingFigure 2
  • EP3278317B1 patent drawingFigure 3

AI summary

A method comprising capturing current and past data frames of a vehicle scenery with an automotive imaging sensor, and predicting, by means of a recurrent neural network, the future position of a moving object in the vehicle scenery based on the current and past data frames.