01 Sep 2025 3 min read Solehound

Project Proposal: Solehound

Problems Identified

When a crime scene is investigated, one of the most common pieces of evidence collected is footwear impressions, known as shoeprints. These impressions are often unaccounted for by suspects and can be valuable evidence for investigators to identify those involved in a crime. However, preforming analysis on footwear impressions is often done manually, making it tedious and cumbersome.

With Machine Learning (ML) seeing rapid progress over the last decade, we felt that this technology could be applied to manual examinations to aid in decreasing processing times for those in the industry. Even though various research studies have proven the feasibility, nobody has yet to attempt to create a product that utilizes ML and is compliant for use in police departments and government agencies.

Project Overview

We are seeking to create a Software as a Service (SaaS) application called Solehound to reverse image search footwear impressions found at crime scenes. This application will utilize Machine Learning (ML) deep learning techniques to identify possible matches for the model of the shoe that created the provided impression. Although it may not initially be admissible evidence in litigation, Solehound can provide useful information for forensic investigators and scientists to identify similarities or possible matches which can significantly reduce the time required. Above all else, it is our primary goal to make applications that are useful for forensic investigators and scientists. Therefore, it is critical to collaborate with those in the industry to ensure the maximum effectiveness of our products.

Project Objectives

The following outlines the goals we wish to achieve by creating a Proof of Concept of this project:

Create a scalable, production-ready full stack application that can be deployed directly to any Docker compatible cloud provider. Ideally, utilizing CI/CD pipelines to automate deployments to test, stage, and production.
If possible, utilize the Arizona Outsole Database that was created by fellow GCU students to further contribute to research efforts. However, if not authorized to use this database, alternative open source datasets such as FID-300 can be used.
Create technical documentation to aid investigators in analyzing impressions, such as Wikis or training materials. Additionally, create example investigations to demonstrate the product’s use case.
Identify and create policies to ensure evidence is admissible and maintains a chain of custody as outlined in the United States Federal Rules of Evidence (FRE).
Ensure the application meets the security requirements outlined by the United States government for independent contractors. For example, CJIS and FIPS compliance.
Implement a proactive security program to attempt compliance with popular security frameworks such as National Institute of Standards and Technology (NIST) Cybersecurity Framework (CSF) and American Institute of Certified Public Accountant (AICPA) SOC 2 .

Project Concerns

When brainstorming possible solutions to tackling this project, a few concerns arose. Below is a list of these initial concerns:

The ML approach must have an efficient method for saving and querying feature vectors. Otherwise, the project cannot handle large data sets.
Maintaining compliance with various laws, regulations, and security frameworks is difficult, timely, and costly.
Cloud provider costs will likely be a concern due to the amount of instances needed, so budgeting will be important.

Project Feedback

We are always open to receiving feedback! If you have a minute, feel free to fill out the survey below to let us know how where we can improve. The survey is anonymous by default, so no sign in or contact information is required. For questions or concerns, please feel free to reach out to us directly at info@kognetiklabs.com.