TensorFlow Object Detection on Raspberry Pi

Real-time object detection system built on TensorFlow Object Detection API, designed to run on Raspberry Pi and webcams. Includes a custom training pipeline for domain-specific models using SSD MobileNet.

!TensorFlow !Python !Platform

Features

• Real-time detection via webcam, video file, or PiCamera
• Custom model training pipeline: annotate → XML → CSV → TFRecord → train → freeze → deploy
• SSD MobileNet v2 architecture for edge deployment
• Pet detection alerts with Twilio SMS integration
• Autonomous robot control via SSH commands on detection events

Architecture

┌─────────────┐     ┌──────────────────┐     ┌─────────────┐
│  PiCamera / │────▶│  TF Object       │────▶│  Actions:   │
│  Webcam     │     │  Detection API   │     │  - SMS alert │
│  Video file │     │  (SSD MobileNet) │     │  - Robot cmd │
└─────────────┘     └──────────────────┘     │  - Logging   │
                                              └─────────────┘

Training Pipeline

# 1. Annotate images with LabelImg → XML annotations
# 2. Convert XML annotations to CSV
python training_pipeline/xml_a_csv.py

# 3. Generate TFRecords for training
python training_pipeline/csv_a_tf.py

# 4. Train with TF Object Detection API
# 5. Export frozen inference graph
# 6. Deploy to Pi
python Object_detection_picamera.py

Files

File	Description
detector.py	Main detector with pet detection, SMS alerts, and robot control
detector1.py / detector2.py	Alternative detector configurations
Object_detection_webcam.py	Webcam-based detection
Object_detection_video.py	Video file detection
Object_detection_picamera.py	Raspberry Pi camera detection
training_pipeline/xml_a_csv.py	Convert XML annotations to CSV
training_pipeline/csv_a_tf.py	Generate TFRecords from CSV
config/	Label maps and model configuration

Requirements

tensorflow>=1.13
opencv-python
numpy
pillow

Year

2018–2020