Source Code Analysis Uncovers AI Patent Infringement

PROBLEM STATEMENT

The Client’s Challenge

A client sought support in investigating potential infringement of a patent covering an AI-based object-detection method.

The patented technology uses neural networks to analyze images, identify objects, generate bounding boxes around detected objects, and classify those objects into categories such as people, vehicles, or other predefined classes.

A key aspect of the invention was the ability to refine the generated bounding boxes so that they more accurately corresponded to the detected objects. This refinement was important because object localization directly affects the reliability of the overall detection process.

The client’s objective was to determine whether another company’s commercial technology implemented the same patented approach.

The investigation focused on the technical functionality associated with:

- Image analysis using AI/neural-network-based object detection.
- Identification of multiple objects within an image.
- Generation of bounding boxes around detected objects.
- Classification of detected objects into corresponding categories.
- Refinement or adjustment of bounding boxes to improve object localization.

CLIENT’S OBJECTIVE

Determine whether the accused product implemented the technical features covered by the client’s patented AI-based object-detection method.

The investigation required more than demonstrating similar product functionality.

The objective was to establish whether the underlying technical implementation corresponded to the patented approach.

TRADITIONAL INVESTIGATION APPROACH

The investigation initially followed a conventional infringement-search methodology focused on identifying commercially available products and technologies that appeared to implement comparable object-detection functionality.

Conventional Product and Technology Search

Product documentation

Technical articles

Marketing materials

Product demonstrations

User reviews

Particular attention was given to indications that a system could adjust, optimize, or refine detection boxes rather than merely draw fixed bounding boxes around detected objects.

Traditional Investigation Workflow

Product Identification

Identify products using AI-based object detection.
Review commercial product descriptions.
Identify systems capable of detecting and classifying objects.

Technical Screening

Review product documentation.
Analyze technical articles and demonstrations.
Examine publicly available screenshots and videos.

Feature Comparison

Determine whether bounding boxes are generated.
Examine whether detected objects are classified.
Look for evidence of bounding-box adjustment or refinement.

Candidate Prioritization

Compare observable functionality with patent requirements.
Identify products showing the strongest apparent technical correspondence.
Select candidates for deeper investigation.

Outcome: Identify commercially available products that potentially practice the patented object-detection functionality and prioritize the strongest candidates for technical analysis.

Limitations of Traditional Approaches

Publicly available product documentation did not provide sufficient implementation-level detail to determine how bounding boxes were generated and refined.
Product demonstrations and user screenshots showed observable detection behavior but could not reveal the underlying object-detection workflow.
Similar bounding-box outputs could result from different technical implementations, making product-level comparison insufficient to establish infringement.
Internal processing associated with bounding-box generation, classification, and refinement was not disclosed through conventional public sources.
The available evidence was therefore stronger for establishing functional similarity than for demonstrating that the accused product implemented the specific patented method.

Start a Conversation →

Wissen’s Out-of-the-Box Approach:

Source Code as the Missing Technical Evidence

Recognizing the limitations of product-level investigation, the strategy evolved beyond publicly observable functionality to examine the underlying implementation of the accused technology.
To enable a more technically grounded infringement assessment, the investigation expanded to include source-code analysis of the relevant product, focusing on the implementation of the patented object-detection workflow.

Identifying Relevant Product Prior Art

A focused technical investigation was conducted on commercially available AI-enabled products demonstrating object detection and classification functionality.
One smart home security product emerged as a particularly relevant candidate because its publicly observable functionality included advanced people and package detection using bounding boxes.
However, publicly available product information did not sufficiently disclose the internal processing used to generate and refine those bounding boxes.
To overcome this limitation, source-code analysis was performed on the identified product.

Our Methodology

Traditional Investigation Limitations

Product-level evidence limitations
Hidden bounding-box implementation
Unobservable refinement logic
Limited internal processing visibility

Identification of Relevant Accused Technology

Screen AI-enabled products
Evaluate detection functionality
Prioritize bounding-box systems

Source Code Acquisition and Review

Obtain available source code
Examine object-detection logic
Identify relevant processing modules

Object-Detection Workflow Analysis

Trace image-processing operations
Identify object detection
Analyze object classification
Examine bounding-box generation

Bounding-Box Refinement Analysis

Trace bounding-box processing
Evaluate refinement logic
Analyze coordinate adjustments
Compare patented refinement approach

Patent-to-Implementation Mapping

Map source-code functionality
Compare patent requirements
Analyze technical correspondence
Document implementation-level evidence

Technical Infringement Assessment

Establish technical correspondence
Corroborate product behavior
Develop infringement evidence

Reverse Engineering Methodology

The source-code analysis focused on the internal implementation of the accused product’s AI-based object-detection workflow to determine whether the patented bounding-box refinement process was implemented.

AI-based object detection methodology image

The source-code analysis revealed implementation-level evidence corresponding to the patented object-detection workflow, particularly the generation, evaluation and refinement of bounding boxes.

Most importantly, the findings established a technical connection between the product’s internal processing and the specific functionality covered by the client’s patent, strengthening the basis for the infringement assessment.

Evidence Collection & Validation

The source-code analysis provided the most important evidence in the investigation, but the findings were considered in conjunction with the product’s observable behavior.

Image Processing

Object Detection

Bounding-Box Generation

Object Classification

Bounding-Box Refinement

The publicly observable product demonstrated object detection and bounding-box functionality, while the source code provided evidence of the internal processing associated with those functions.

Technical Challenges Addressed

Challenge 1: Determining How Bounding Boxes Were Generated

The first technical question was whether the accused product merely displayed bounding boxes or whether its underlying implementation generated those boxes through a process relevant to the client’s patent.

Solution

Source-code analysis was used to trace the object-detection workflow from image processing through bounding-box generation.

The analysis revealed:

Image preprocessing workflows
Object detection processing logic
Bounding-box generation mechanisms
Detection coordinate processing
Object localization parameters

The findings established implementation-level evidence corresponding to the object-detection and bounding-box generation aspects of the patented technology.

Challenge 2: Establishing Correspondence Between Product Implementation and Patent Requirements

Even after identifying object detection and bounding-box refinement functionality, the investigation needed to determine whether the overall implementation corresponded to the technical method claimed by the client’s patent.

Solution

The identified source-code operations were compared against the relevant patent requirements.

The analysis revealed:

Bounding-box refinement mechanisms
Object classification processes
Sequential processing relationships
Patent limitation correspondence
Implementation-level technical evidence

This implementation-level comparison demonstrated substantial correspondence between the accused product’s processing approach and the patented object-detection method.

Results & Business Impact

The analysis revealed that the accused product’s object-detection workflow included processing associated with generating, evaluating, and refining bounding boxes in a manner closely corresponding to the client’s patented method.

Key Outcomes

Identification of a commercially available AI-based object-detection product relevant to the patent investigation.
Establishment of implementation-level evidence through source-code analysis.
Identification of object-detection and bounding-box generation workflows.
Identification of processing associated with bounding-box refinement.
Technical comparison of the implementation against the client’s patented method.
Development of evidence supporting the client’s patent enforcement strategy.

Strategic Business Impact

By moving beyond conventional product-level comparison and incorporating source-code analysis, the investigation provided:

Stronger technical support for infringement assessment
Improved visibility into proprietary implementation details
Greater confidence in mapping product functionality to patent requirements
More focused evidence for potential licensing or enforcement discussions
Broader investigative coverage beyond publicly observable product behavior

The approach demonstrated how combining patent claim analysis with implementation-level software investigation can uncover technical evidence that may remain inaccessible through conventional product research alone.

Conclusion & Implications

This case study highlights the importance of adopting deeper and more adaptive investigation strategies when assessing potential infringement of AI and software patents.
Traditional product and technology research remains valuable for identifying potentially relevant systems and establishing observable functionality. However, commercial AI products often expose only their outputs, while the underlying processing logic remains inaccessible through conventional public sources.
By expanding the investigation from product-level analysis to source-code-level examination, the project enabled a more technically grounded assessment of how the accused system generated, evaluated, and refined bounding boxes. This provided evidence that could be directly compared with the technical approach described in the client’s patent.
Ultimately, the investigation demonstrated how source-code analysis can significantly strengthen infringement assessments when the critical patent limitations relate to an underlying software workflow rather than simply to the final product functionality.
The investigation also reinforced the importance of:

Looking beyond observable product behavior

Examining the underlying software workflow

Mapping technical implementation to patent requirements

Using source-code evidence

Adapting the investigation methodology

Could Source Code Reveal Infringement Hidden Behind Imperfect AI Object Detection?