Neural Network Output Stabilization Through Multi-Object Tracking

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Neural network-based object detection in video streams is unstable due to variations in lighting and perspective, leading to inconsistent results over time, and existing methods to enhance stability, such as predicting images multiple times, are computationally expensive.

Innovation Solution

A computer-implemented method that stabilizes neural network outputs by associating data elements with unique identifiers, determining contextual information, and generating outputs based on historical data to reduce fluctuations, using averages and weighted probabilities to track objects over time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If frame-by-frame object detection is applied to each video frame individually, then detection speed is maintained, but output stability deteriorates due to variations in lighting and perspective

Engineering Contradiction:
Improvedetection speedVSAvoidoutput stability
Core Design Contradiction:
SpeedVSStability of the object's composition

Solution Approach 1:

The system performs preliminary actions by maintaining historical detection results and contextual information from previous frames. Before generating the current frame's output, the system prepares tracking data, historical classifications, and contextual features that are used to stabilize the final output, preventing fluctuations caused by individual frame variations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements feedback mechanisms by continuously comparing current detection results with historical tracking data and contextual information. The output for each frame is refined based on feedback from previous detections, ensuring consistency across frames while maintaining detection speed through efficient feedback integration.

Inventive Principle:
Principle #23Feedback

2Measurement precision

If multiple predictions are performed on the same image to estimate uncertainty, then uncertainty estimation accuracy is improved, but computational cost increases significantly

Engineering Contradiction:
Improveuncertainty estimation accuracyVSAvoidcomputational cost
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system introduces contextual information as an intermediary that mediates between the neural network output and the final detection result. This contextual information includes historical tracking data, temporal patterns, and environmental context that help estimate uncertainty without requiring multiple redundant predictions, thereby reducing computational cost while maintaining estimation accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

Instead of performing multiple full predictions on the same image, the system creates simplified copies or representations of the detection problem using contextual information from previous frames. These contextual copies provide uncertainty estimates without requiring the full computational expense of multiple independent predictions.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4600914A1Neural network output stabilization and uncertainty estimation through multi-object-tracking
Publication Date: 2025.08.13 HELSING GMBH
  • EP4600914A1 patent drawingFigure 1
  • EP4600914A1 patent drawingFigure 2
  • EP4600914A1 patent drawingFigure 3

AI summary

According to an aspect of the present disclosure, there is provided a computer-implemented method, in particular for object tracking, the method comprising: receiving a stream of data items containing one or more data elements; associating a first data element of the one or more data elements with a unique identifier; determining at least one feature associated with the first data element; and for each feature: determining contextual information relating to the feature; and generating an output indicative of the feature based on the contextual information, wherein the output and said identifiers are useable for tracking an object represented by said first data element. In an embodiment, the output includes an average of values included in said contextual information. Determining an average enables generating a more stable output and reducing the amount of data that needs to be processed to generate the output. Thus, the method enables stabilizing the output of a neural network by tracking and averaging said values.