Project 4: Real-Time Object Counting & Tracking

Build a people/vehicle counter for retail analytics, traffic monitoring, or warehouse management.

What You’ll Learn

  1. Object Detection - Using YOLO for real-time detection
  2. Multi-Object Tracking - Centroid-based tracking algorithm
  3. Counting Logic - Line crossing detection
  4. Direction Detection - In/out movement tracking
  5. Analytics - Generating traffic statistics

Key Concepts

Concept Description
Detection Finding objects in each frame
Tracking Maintaining object identity across frames
Counting Line Virtual line for crossing detection
Centroid Center point of bounding box

Usage 

# Run demo with simulated objects
python main.py --demo

# Use webcam
python main.py --camera

# Process video file
python main.py --video traffic.mp4

# Count specific classes
python main.py --camera --classes person car

Algorithm 

1. Detect objects in frame (YOLO)
        |
2. Extract centroids
        |
3. Match with existing tracks
   (Hungarian algorithm / nearest neighbor)
        |
4. Update track positions
        |
5. Check line crossing
        |
6. Update counts

Tracking Algorithm

Simple centroid tracker:

  1. Get centroids from new detections
  2. Compare with existing object positions
  3. Match based on minimum distance
  4. Handle new objects and disappeared objects

Real-World Applications

  • Retail foot traffic analysis
  • Traffic monitoring
  • Warehouse inventory tracking
  • Crowd management
  • Parking lot occupancy

Code Highlights

Centroid Matching 

# Compute distances between existing and new centroids
for i, obj_c in enumerate(object_centroids):
    for j, inp_c in enumerate(input_centroids):
        D[i, j] = np.sqrt((obj_c[0] - inp_c[0])**2 +
                          (obj_c[1] - inp_c[1])**2)

Line Crossing Detection 

# Check if object crossed the line
if prev[1] < line_y <= centroid[1]:
    count_down += 1  # Crossed downward
elif prev[1] > line_y >= centroid[1]:
    count_up += 1    # Crossed upward

References