These days, it's easy to automatically find the shoes you want on shopping websites just by uploading a photo. These automatic retrieval algorithms rely on shoe color, surface features, and other characteristics to locate matching products. But what about identifying a shoe from a shoeprint left on the ground? How can we accurately determine which shoe left that trace? Today, let me introduce our latest advancements in this area.
Our previous generation of shoeprint retrieval algorithms relied on manually outlining tread patterns and extracting their features. These extracted features were then used to search a database for shoes with matching outsoles.
This method had its advantages: by continuously refining the tread pattern features, it was possible to eventually find a matching shoe. However, it also had clear drawbacks:
1. For new users unfamiliar with the software, learning to draw shoeprint features had a steep learning curve.
2. Different individuals may interpret the same shoeprint pattern differently, leading to ambiguity in the extracted features.
To solve these problems, our technical team conducted years of research and large-scale data analysis, resulting in the development of a third-generation AI-based automatic shoeprint recognition algorithm. With this algorithm, there's no need to manually process tread features — you simply mark the area ofthe outsole in the image, and the system will handle the rest.
Just as technologies like facial recognition, gait analysis, palm and fingerprint recognition, DNA, voiceprint, and iris recognition are advancing in the field of forensic science, we have dedicated the past 25 years to researching footwear evidence, which is the most commonly left trace at crime scenes. We continue to upgrade our matching algorithms and develop new applications. We sincerely hope our work can support forensic experts and law enforcement officers who rely on footwear evidence to solve cases.
Note: The shoeprint used in the video was collected usingthe professional on-site shoeprint collection device EverISS and has not been manually processed.
.png){kind=small}
