Skip to content

Case study

AI-Powered Rod and Stem Detection and Counting System Delivering 98%+ Counting Accuracy on Manufacturing Conveyor Lines

MoreYeahs built an AI-powered vision system combining a custom-trained YOLOv8 segmentation model with ByteTrack multi-object tracking to detect, segment, classify, and count rods and stems in real time.

2 min readManufacturingAI and Machine Learning

Key result93%+ detection and classification accuracy achieved for rods and stems.
AI and Machine Learning
Industry
Manufacturing
Engagement
AI-Powered Vision-Based Production Counting
Focus
Real-Time Rod and Stem Detection and Counting

Results at a glance

  1. 0193%+ detection and classification accuracy achieved for rods and stems.
  2. 0298%+ counting accuracy achieved compared to ground truth.
  3. 03Duplicate count rate held at 2% or below, with tracking ID stability of 95%+.

01The client

Customer

Manufacturing and processing facilities face challenges accurately detecting, classifying, and counting rods and stems moving on conveyor belts, where manual inspection is slow, inconsistent, and prone to error.

MoreYeahs built an AI-powered rod and stem detection and counting system for manufacturing organizations, giving production and quality control teams accurate, real-time counts without manual tallying.

02Challenge

Manual counting and legacy vision systems could not keep pace with high-speed, high-similarity production flow:

  1. 01Slow, Error-Prone Manual Counting

    Manual inspection and counting were labor-intensive and inconsistent between operators.

  2. 02Similar-Looking Objects

    Traditional sensor-based or rule-based vision systems struggled to distinguish similar-looking rods and stems.

  3. 03Overlap and High-Speed Flow

    Partial overlap and continuous high-speed conveyor movement caused missed or duplicate counts.

  4. 04Limited Production Visibility

    Lack of accurate real-time counts reduced visibility for quality control and throughput decisions.

03Solution

Solution

MoreYeahs built an AI-powered vision system combining a custom-trained YOLOv8 segmentation model with ByteTrack multi-object tracking to detect, segment, classify, and count rods and stems in real time.

Pixel-Level Segmentation

YOLOv8 segmentation ensures precise object boundary detection for each rod and stem.

Persistent Object Tracking

ByteTrack assigns a persistent tracking ID to every detected object across frames.

Virtual Counting Line

A configurable line automatically increments class-wise counts when tracked objects cross it in the correct direction.

Operator Interface

A Streamlit-based web interface lets operators upload video, tune detection parameters, set the counting line, and export annotated video and count statistics.

04How we delivered

The pipeline runs from video ingestion through segmentation, tracking, and counting, with safeguards against duplicate counts.

  1. Step 1Segmentation and Classification

    YOLOv8 generates pixel masks and class labels for each rod and stem in the frame.

  2. Step 2Trajectory Maintenance

    ByteTrack maintains object trajectories across frames, with tracking ID history preventing recounts.

  3. Step 3Color-Coded Visualization

    Color-coded overlays distinguish rods and stems, with a dashboard showing real-time class-wise counts and totals.

  4. Step 4Resilient Handling

    Overlapping or clustered objects are tracked as a group until separation, and temporarily lost detections retain their ID if re-detected within a buffer window.

05Technology

The solution combines an industrial conveyor camera setup with a custom segmentation and tracking stack.

YOLOv8 Segmentation Model
Custom-trained for rod and stem detection and classification
ByteTrack Multi-Object Tracker
Persistent ID assignment for accurateduplicate-free counting
GPU-Enabled Processing
Hardware sized for real-time conveyor-speed throughput
Streamlit Interface & Export Pipeline
Video I/Oannotationand CSV/video statistics export

06Results

In validation testing, the system met its target benchmarks for detection, counting, and tracking accuracy.

  • 93%+ detection and classification accuracy…

    93%+ detection and classification accuracy achieved for rods and stems.

  • 98%+ counting accuracy achieved compared…

    98%+ counting accuracy achieved compared to ground truth.

  • Duplicate count rate held at…

    Duplicate count rate held at 2% or below, with tracking ID stability of 95%+.

  • Real-time processing sustained at required…

    Real-time processing sustained at required conveyor speed, with annotated video export working reliably.

07Business impact

Accurate, automated counting gives manufacturing teams real-time production visibility without manual tallying.

Reliable Production Counts

Near-zero duplicate counting gives quality and operations teams a trustworthy throughput number.

Reduced Manual Labor

Automated counting frees staff from repetitive manual tallying.

Operator Self-Service

The tunable interface lets line operators adjust detection parameters and counting rules without engineering support.

Extensible to Other Lines

The segmentation-and-tracking approach can extend to other similar-object counting use cases in the plant.

More projects like this

All case studies

Facing a similar challenge?

Tell us what you need. A technical lead will reply within one business day with how we would approach it.

  • Reply within one business day
  • Free 30-minute call
  • NDA before you share data