Pedestrian-Detection

Introduction

This is the final project of the Udacity Machine Learning Engineer Nanodegree Program.

As final project, i created a simple pedestrian detection system, that useses an image classifier and a sliding window to detect pedestrians in an given input image. The dataset used in this project is from the Penn-Fudan Database for Pedestrian Detection and Segmentation. For the classifier, the resnet50 architecture is adapted (feature extraction on final fully conected layer).

Included in this repository

• Data Exploration and Generation.ipynb - get familiar with the data and generate classifier input images
• The Classifier.ipynb - setting up and traning of the classifier
• Pedestrian Detection Pipeline.ipynb - End to end usage of the created system on examples
• Benchmark Model and Evaluation.ipynb - Setup and evaluation of both models
• DataClass.py - Contains PennFudanDataset class that is useful to load the data and work with it
• detection.py - Outsourced detect function of the produced model (for usage across notebooks)
• helpers.py - useful helper fuctions (such as sliding window, image pyramid, display bounding boxes, etc.)
• model.py - detection model with tranined weights

Setting up the environment

Dataset

The zip-file has to be downloaded manually. To unzip the files the first notebook can be used.

Libraries

This project is developed in Python 3.6 You will need to install some libraries in order to run the code.
Libraries and respective version are:

• jupyter 1.0.0
• opencv-python
• numpy 1.18.1
• pandas 1.0.1
• pathlib2 2.3.5
• Pillow 7.0.0
• zipp 2.2.0
• matplotlib 3.1.3
• torch 1.4.0
• torchvision 0.5.0
• glob2 0.7
• natsort 7.0.1
• imutils 0.5.1
• split-folders 0.4.2