Defect detection for finding defects on objects, then identifying and tracking them. The solution processes visual information and monitors it consistently to support quality control and operational decisions.
Overview
Defect Search is a visual quality-control solution for retail. It finds defects on objects, identifies them, and keeps track of each object while it stays in view.
The system reads visual information and provides ongoing monitoring, so quality checks and operational decisions can rely on a consistent signal rather than a one-off look.
The model was built on TensorFlow and optimized to run on Nvidia Jetson, covering 12 types of defects.
The Challenge
Objects on a line can show defects that need to be found and named. A single glance is not enough when the same object stays in the camera’s field of view and would otherwise be reported again.
The monitoring also had to stay consistent enough for quality control and for people making operational decisions from that visual feed.
Our Solution
We built a model that searches for defects, identifies the type, and tracks the object while it remains in view so the same defect is not signaled twice.
That model is based on TensorFlow and was optimized to run on Nvidia Jetson, close to where the visual information is captured.
Key Features
The system finds defects on objects and identifies which kind of defect it is seeing.
An object is tracked while it stays in the field of view, so the same defect does not raise a duplicate signal.
The Impact
The team created its own TensorFlow model, optimized it to run on Nvidia Jetson, and trained it to recognize 12 types of defects.
12
Types of defects
TensorFlow
Own model
Jetson
Optimized for Nvidia Jetson