Object Tracking via Time Frame Segmentation for Embedded Speed Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object tracking systems in driver assistance and autonomous driving face challenges in processing resource limitations and real-time speed detection, particularly with techniques relying on machine learning and neural networks, which are computationally expensive and resource-intensive, especially in embedded systems with limited memory bandwidth.

Innovation Solution

A method that splits time frames into sub-intervals to track object motion using convolutional recurrent neural networks, aligning previous and current data through standard convolutions and sub-intervals, allowing for efficient tracking and speed estimation within the capabilities of embedded systems, avoiding irregular memory accesses and relying on Doppler speed information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning and neural network techniques are used for speed detection and object tracking, then detection accuracy and tracking precision are improved, but processing resource requirements increase beyond capabilities provided by processing resources in vehicles

Engineering Contradiction:
Improvespeed detection accuracyVSAvoidprocessing resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The time frame is split into multiple sub-intervals, allowing the system to process motion data in smaller, manageable segments. This segmentation enables the tracking system to handle complex motion patterns without requiring excessive computational resources, as each sub-interval can be processed independently with standard convolutional operations rather than requiring complex neural network inference.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent uses standard convolutional operations that replicate and shift previous time frame data to create predictions for the current time frame. This copying approach replaces complex neural network processing with efficient convolutional operations that can be performed on embedded systems, maintaining tracking accuracy while reducing computational burden.

Inventive Principle:
Principle #26Copying

2Measurement precision

If machine learning and neural network techniques are used for speed detection and object tracking, then detection accuracy and tracking precision are improved, but device complexity increases

Engineering Contradiction:
Improvetracking precisionVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces complex neural network algorithms with standard convolutional operations that can be implemented using traditional image processing techniques. This substitution simplifies the system architecture by using well-established convolutional building blocks rather than complex neural network models, making the system more suitable for embedded vehicle processors.

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

3Loss of time

If object tracking is performed in real-time for dangerous situation detection, then response time is improved, but processing resource consumption increases

Engineering Contradiction:
Improveresponse timeVSAvoidprocessing resource consumption
Core Design Contradiction:
Loss of timeVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary processing by splitting the time frame into sub-intervals and preparing motion data in advance. This preliminary action allows the tracking system to maintain real-time performance by having pre-processed data ready for immediate analysis, reducing the computational burden during critical real-time detection moments.

Inventive Principle:
Principle #10Preliminary action

4Use of energy by moving object

If standard convolutional operations are used instead of neural networks, then processing resource requirements are reduced, but tracking capability for motions exceeding receptive field size is limited

Engineering Contradiction:
Improveprocessing resource consumptionVSAvoidtracking capability
Core Design Contradiction:
Use of energy by moving objectVSAdaptability or versatility

Solution Approach 1:

By dividing the time frame into multiple sub-intervals, the system enables standard convolutional operations to track objects with motions that exceed the receptive field size of individual convolutional kernels. The segmentation allows cumulative tracking across multiple sub-intervals, effectively extending the tracking capability without requiring larger receptive fields or more complex neural networks.

Inventive Principle:
Principle #1Segmentation

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach enables accurate and efficient tracking of objects with motions exceeding the receptive field size of standard tracking systems, reducing computational costs and memory requirements, thus improving real-time object tracking and speed detection in resource-constrained environments.

Implementation Method 1

relying on Doppler speed information

Methodology Applied
Scientific EffectDoppler effect: Doppler Effect

Data Source

PatentEP4379659A1Enhanced tracking and speed detection
Publication Date: 2024.06.05 APTIV TECHNOLOGIES AG
  • EP4379659A1 patent drawingFigure 1
  • EP4379659A1 patent drawingFigure 2
  • EP4379659A1 patent drawingFigure 3A

AI summary

The subject disclosure pertains to object tracking for use in advanced driving assistance systems and autonomous driving systems of vehicles. For example, the subject disclosure provides a computer-implemented method for tracking an object. The method comprises obtaining a motion of the object within a time frame based on data received from a sensing system, splitting the time frame into a plurality of sub-intervals, and determining, using a tracking system, a position of the object in a next time frame based on sub-motions of the object within the plurality of sub-intervals.