Днище автомобиля
Автомобили
ISS logo

A system for detecting foreign objects on the underside of cars using a neural network

Image analysis

Basic information

Customer

ISS - Intelligent Security Systems - company-integrator of security solutions
ISS logo

Product

Task

Software for test benches of a car - the underside is inspected by a person using mirrors and photographs
Find abnormal objects at the bottom of the car using a photo

Where is used:

at military sites

Why it is needed:

at border posts
in places with increased safety requirements
automate the verification process
speed up the operator's work

How it works

The car arrives at the stand
Посторонний предмет на днище автомобиля
Photo-shooting of the car underside is being carried out
When anomalies are found, it shows the snapshot to the operator
The neural network analyzes the snapshot
Anomaly

Technical implementation

Conducted ML audit

Since such tasks were almost not solvable, our client ordered an ML-audit service

Found a scientific article

As a result of the research, we found a scientific article on a similar topic.

Model

We reproduced the article and adapted the model to our data.

VisionHub

As a visualization interface, we used our hosting of VisionHub machine vision models.
Модель данных
Хостинг моделей машинного зрения VisionHub

GAN

The main method is generative adversarial networks (GAN), a machine learning algorithm.
Principle of operation:
Step 01
Remember how an object looks like in a normal state
Step 02
Generate images and capture anomalies on them
Объекты в нормальном состоянии
Зафиксирован аномальный объект
Anomaly detected

Development process

Metric - the ratio of correctly identified anomalies and falsely identified

Anomaly: yes / no on this fragment. There was no task to accurately localize.

Input data

20
6
images of the underside of cars with anomalies
reference images of the bottom
Very small dataset, insufficient for high-quality neural network training
High resolution images - 4K by 10K pixels

Dataset processing

24
317
We manually marked the images, noted the anomalies.
images with anomaly
normal
Днище автомобиля
We cut the photos into 512-pixel squares and looked at them separately.
130
3221
abnormal
normal

Model training

Trained the model on a prepared sample.

The accuracy of the quality metric was 0.73

Development stages

Harmonization
Audit
Start of development
1.5-2 months
2 weeks

Current results

The audit stage has ended
Managed to build the first model
Getting ready for a big project

Project team

Manager
ISS logo
Mikhail Evdokimov
Techlead
Vlad Vinogradov
Computer Vision Engineer
Aelita Shaikhutdinova

Features of the project

First used GAN in production
One of our first audits
This is the first time we work with an integrator company and, in general, with people who are very indirectly related to the development.

Where groundwork can come in handy

For example, in medical applications
In systems where deviation from the norm needs to be detected

Difficulties of the project

Unusual material
Communication difficulties, since the company is very large
Contact us