Аннотации:
The traditional approach to criminal investigation,
taken in the context of rapidly developing technological advancements,
presents itself to be rather outdated. One of the
most crucial parts of identifying a suspect is criminal sketching.
The modern process requires manual sketch craftsmanship from
assets based on the description of the suspect in order to
approximate the appearance of the culprit. This method is highly
limited and burdensome, it requires a considerable amount of
time and resources that could be used in other areas of law
enforcement.
The proposed system utilizes machine learning techniques
to approach this problem and suggests an alternative way of
identifying criminals. The system analyzes photos uploaded to
the database and evaluates pre-defined metrics to synthesize a
vector for a person that will be used to describe the appearance in
abstract terms. The system then takes prompts from the intended
users, which could be victims or witnesses, to synthesize a new
vector for each prompt and will find the most corresponding
photos from the database.
The results of testing the product indicate decent accuracy
and usability and propose a direct search from the database of
people instead of a traditional empirical search from a manual
approximated sketch. This method will remove the necessity for
the time-consuming manual sketching process and further search
for an imprecise image of a suspect. A simple and intuitive in-use
interface will not require any special training of the police staff
in order to integrate the product into the current system of law
enforcement.
The project might be limited as of today, but given enough
time and resources, several additions will drastically improve
the effectiveness of criminal investigation and contribute to the
redistribution of the police force on other important tasks.