Multi-Scale CNN Object Detection With Image Shifting Compensation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Multi-scale CNNs in object detection suffer from lower confidence in detecting objects at certain horizontal or vertical locations due to the scaling process, affecting the interpretation of object extent, leading to incorrect detections.

Innovation Solution

An object detection arrangement that utilizes a multi-scale convolutional neural network to classify potential detection errors based on object location and adapt detection parameters or shift the image to improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multi-scale CNN is used for object detection, then detection capability for objects of different sizes is improved, but detection confidence at certain locations deteriorates due to scaling artifacts

Engineering Contradiction:

Improvedetection capability for objects of different sizesVSAvoiddetection confidence at certain locations

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The system performs preliminary analysis to identify objects located at critical distances (multiples of confidence distance) before final detection. By detecting these at-risk objects early and applying compensation measures such as adjusting detection thresholds or performing additional verification, the system prevents location-related confidence losses from affecting final detection reliability.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts detection parameters based on object location. When an object is detected at a critical distance where scaling artifacts reduce confidence, the system modifies detection thresholds, confidence requirements, or processing parameters to compensate for the location-induced confidence loss, thereby maintaining overall detection reliability across all locations.

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If multi-scale CNN downsampling is applied, then multi-scale feature extraction is improved, but measurement precision of object extent deteriorates at specific distances

Engineering Contradiction:

Improvemulti-scale feature extractionVSAvoidprecision of object extent interpretation

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system implements feedback mechanisms that monitor detection confidence levels and object location metrics. When objects at critical distances are detected, the feedback loop triggers compensatory actions such as adjusting detection thresholds or performing additional verification passes, which corrects the measurement precision errors introduced by downsampling at those specific locations.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

Before final object extent measurement, the system performs preliminary identification of objects at critical distances. This preliminary action allows the system to apply enhanced processing or adjusted measurement criteria specifically for these objects, compensating for the precision loss caused by multi-scale downsampling artifacts.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12602905B2Computer software module arrangement, a circuitry arrangement, an arrangement and a method for improved object detection adapting the detection through shifting the image

Publication Date: 2026.04.14 TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)

AI summary

An object detection arrangement (100) comprising a controller (101) configured to detect objects utilizing a multi-scale convolutional neural network, wherein the controller (101) is further configured to: receive (312, 410) image data representing an image (10) comprising an object to be detected (11) being at a distance (d) into the image (10); classify (314, 430) whether the object to be detected (11) is at risk of being incorrectly detected based on the distance (d); and if so compensate (315, 440) the object detection by shifting (317) the image (10).