Mouse Pellet Detection
We developed a comprehensive video analysis solution to track mouse movement, count pellets, and classify them by color in a controlled research environment. The system leverages advanced object detection, motion tracking, and color segmentation techniques to provide accurate, timestamped behavioral data for scientific analysis.
What We Built
A fully automated video analysis pipeline using:
- YOLO for object detection of mice and pellets
- ByteTrack for continuous multi-object tracking of mouse movements
- SAM 2 (Segment Anything Model) for pixel-level pellet segmentation, even when partially occluded
- Custom color detection algorithm for red vs. non-red pellet classification
- Make Sense for video frame annotation
- Roboflow for dataset preparation, labeling, and augmentation
- Google Colab Pro for GPU-enabled model training and video processing
The system acts as a precision behavioral analysis engine, dynamically processing video footage to extract quantitative data on mouse behavior and pellet interactions.
Key Features
Data Annotation
- 10,000 diverse frames annotated across 20 videos
- Robust training set for accurate detection
Mouse Detection & Tracking
- Real-time identification of mice in video frames
- Continuous tracking using ByteTrack for movement analysis
- Motion pattern detection and distance calculations
Pellet Segmentation
- Pixel-level segmentation using SAM 2 for occluded pellets
- Accurate counting of pellets dropped after video start
Color Detection
- Custom algorithm classifying pellets as red or non-red
- Threshold-based classification for consistent results
Timestamp Detection
- Precise logging of when pellets fall
- Essential for behavioral timing analysis
Motion Analysis
- Pattern detection of pellet drops and mouse movements
- Comprehensive behavioral insights
How It Works
Video footage is ingested and frames are extracted
YOLO detects mice and pellets within each frame
SAM 2 performs pixel-level segmentation for occluded pellets
ByteTrack tracks mouse movement continuously across frames
Color detection algorithm classifies pellets by color
Timestamped data is compiled into comprehensive CSV reports
Results
96%
detection accuracy for pellet identification
Completed
within 5-week timeline with consistent milestone delivery
18%
reduction in false negatives
Accurate
differentiation between red and non-red pellets
Detailed
timestamped data for behavioral analysis
Value Dilevered
This project transformed manual video analysis into an automated, high-precision behavioral tracking system. By combining object detection, motion tracking, and color segmentation, the solution provides researchers with accurate, timestamped data essential for understanding mouse behavior streamlining research processes and enabling more robust scientific conclusions.
Want to build AI-powered Problem solutions?
#Computer Vision #Object Detection #YOLO #ByteTrack #SAM 2 #Behavioral Analysis #Video Processing #Research Automation