tla001 / OnlineDetector

Online Object Detection System Based On Deep Learning

Online Object Detection System Based On Deep Learning

• This project is a system for objects detection based on deep learning(darknet --yolo),and includes some parts:

1.video stream detect

use ffmpeg,perform that:

• 1. get video stream from camera/file/net video stream
• 2. decode and display with opencv
• 3. encode video with h264 in flv format
• 4. push video through rtmp protocol to nginx stream media server or save as file

the main work of ./client/ObjectDetector:

• pusher: get video from camera-->decode and display with opencv -->encode in h264as flv --> push stream to nginx server
• puller: get video from nginx server-->decode and display with opencv -->encode in h264as flv --> save as file

the main work of ./server/video

get video from nginx server--> decode and dislay with opencv --> detect whih yolo -->display detect results --> encode in h264 as flv --> push stream to nginx server(another channel)

2.image detect

tranport image or json with http,using libevent to listen http in server,using libcurl as client

the main work of ./client/imagedemo/curlclient

read a image file --> transport in http with libcurl ----waiting for reback-- get json stream -->decode json

the main work of ./server/image

libevent as http server in child process--- wait for client --- get a request --> write image stream as file --> signal to main process and yolo read file and detect -->write detect results in a json file --> fifo to wake up child process --> read json file and sent to client with http

CAUSTION:

when use codes in server folder,you must configure darknet already,then you can put these files in src,and replace default Makefile with my Makefile.Also, you have to configure libs I have used.

Something maybe more clear in my blog:www.cnblogs.com/tla001