cv_potholes_detection

Detect potholes in images using Deep Learning technique (YoloV4 AlexeyAB/darknet)

Getting started

Build the docker container.

sudo docker build -f Dockerfile -t docker-cv-pothole-detect .

Download the model weights and move it into models/ folder.

YoloV4 - https://drive.google.com/open?id=1-4BRAxU-ijkp6MlxQ2maetRrnpQ9DjTF
YoloV3-tiny-prn - https://drive.google.com/open?id=10-mUlpiTl2T69LBI7vp-hXkO-Z5g-ZH1.

Setting up env

Copy env file and rename it in .env. Change variables in .env with your custom values. Ex.

# Http port of micro-service
HOST_PORT=5000

# Http url to confirm the completion of job
CALLBACK_URL=http://httpbin.org/post

# Folder in which are stored image's files attached to pothole events
INPUT_IMGS_DIR=/var/www/MCPS/cv_potholes_detection/frames_in

# Images generated by detection with objects bounding box
OUTPUT_IMGS_DIR=/var/www/MCPS/cv_potholes_detection/frames_out

# Host path of model's file directory
MODEL_DIR=/var/www/MCPS/cv_potholes_detection/models

# Model's architecture configuration file
MODEL_CONFIG=models/yolov4-spp-pothole-test.cfg
# or models/yolov3-tiny-prn-pothole-test.cfg for tiny version

# Model's pretained weights
MODEL_WEIGHT=models/yolov4-spp-pothole-train_7000.weights

# Discard detected objects with probability less than 
THRESHOLD=0.55

Run

Launch the service using:

sh run_docker.sh

Endpoints

[POST] /analyze
|Params|: List of pothole events in json format