Skip to content

CNN Object Detection (MobileNet) , Dehazing (AOD-NET) on Caffe,UP Square Board, Movidus Neural Compute Stick.

Notifications You must be signed in to change notification settings

ac-optimus/ESDC_IntelCup2018

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 

Repository files navigation

image9

ESDC_IntelCup2018

The project was presented in ESDC_IntelCup-2018 held at SJTU,Shanghai,China. The competition theme was AI and IOT, with that in mind the project provided a solution that could assist fire-fighters in after fire situation (when the fire is out).
Many fire-fighters death occur every year due to fire. Our solution Dehazes and cleans the input frames for higher level computer vision task like person-detection. Apart from that our solution will give the 3-d map of the location for better judging the harm caused at the rescue site. To detect potential person alive victims on the resuce site motion magnification is employed on the person detected frame. All the details methioned above are live streamed to the rescue site using FLASK (Python Library) server on the local area network.

Intelligent Rescue Operation Bot (IROB) is the implementation of the idea that we propose. As could be seen in the image(below) the bot has Kinetic V2 Xbox One(Depth camera), RGB-Camera, Intel Movidius NCS (inference stick provide hardware acceleration for deep learning models on embedded platform)and Intel Up-Square as main hardwares apart from other peripherals.

This repository only contains the frame processing part where the frames for the resuce site is dehazed and then person detection is applied on it.

For dehazing and person-detection we used the state of art algorithms. The RGB image provided by the RGB camera could process dehazing and person-detection on UP-Square board. Both dehazing and person-detection was implemented using CNN approach (All the details related to the model used could be found in the link below). We used pre-trained models to prototype the idea. For dehazing IROB uses AOD-Net and for person-detection MobileNet_SSD (Integration of MobilNet and SSD) is used. MobileNet suits the purpose better here than the other state of art methods like YOLO due to its light weight and better compatibility with embedded platforms. Intel Movidus NCS does the inferencing to perform the person detection task.

IROB (Intelligent Rescue Operation Bot)

1

2

For more details and Algorithms

AOD-Net(All-in-One Dehazing Network) - https://ieeexplore.ieee.org/document/8237773/

MobileNet - https://arxiv.org/abs/1704.04861

SSD (Single Shot MultiBox Detector) - https://www.cs.unc.edu/~wliu/papers/ssd.pdf

Intel Movidius NCS - https://developer.movidius.com/

Up-Square Board - http://www.up-board.org/upsquared/

YOLO - https://pjreddie.com/darknet/yolo/

About

CNN Object Detection (MobileNet) , Dehazing (AOD-NET) on Caffe,UP Square Board, Movidus Neural Compute Stick.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published