Image Based Object Detection System for Self-driving Cars Application

This project is based on YOLO V1 & YOLO V2
This repository used code from YOLO V2
Presentation Slides can be found here

Introduction

Competition
This project is to perform detection on 4 classes of objects (Vehicle, Pedestrian, Cyclist and Traffict Lights) on the road based on 10,000 training image. The model uses YOLO V1 MXNET and YOLO V2 Darknet framework.

Result

For YOLO V1 please check demo_test.ipynb. I ran 600 epochs and here is detect_full_scale-0600.params.
For YOLO V2 please run following command on test image:

./darknet detector test cfg/obj.data cfg/capstone.cfg backup/capstone_40000.weights testing/70495.jpg

alternatively, you can run this command to convert all validation set and store results in results/ folder:

./darknet detector valid cfg/obj.data cfg/capstone.cfg backup/capstone_4
0000.weights -i 0

This can be used to get final output and calculate mean AP.

Thanks to John, we have two great test videos:
Downtown Video
Highway Video
And run the following commands to check:

cd darknet_alex/darknet
./darknet detector demo cfg/obj.data cfg/capstone.cfg ../../yolo\ v2/darknet/backup/capstone_40000.weights ~/Downloads/Driving\ Downtown\ -\ Toronto’s\ Main\ Street\ -\ Toronto\ Canada.mp4 -out_filename downtown.avi

./darknet detector demo cfg/obj.data cfg/capstone.cfg ../../yolo\ v2/darknet/backup/capstone_40000.weights ~/Downloads/Highway\ 401\ Through\ Toronto\ Worlds\ Busiest\ Freeway.mp4 -out_filename highway.avi

Installation

1. For YOLO V2 please build darknet:

cd darknet
make

Citing

@article{redmon2016yolo9000,
  title={YOLO9000: Better, Faster, Stronger},
  author={Redmon, Joseph and Farhadi, Ali},
  journal={arXiv preprint arXiv:1612.08242},
  year={2016}
}