This code is the implementation of Yolov3 This object detection technique is being used widely in many industrial sectors. It can be said that yolov3 is a competitor of SSD. Yolov3 can be used where speed is an important criteria. On the other hand, SSD can be used where accuracy is the main criteria. Earlier methods,(R-CNN, Fast R-CNN), a sliding window tried to locate objects in an image which is quite time consuming. Faster R-CNN uses Region Proposal Network (RPN) to identify bouding boxes. Yolov3 takes a completely different approach towards object detection. First it divides the image into grid of cells. Each cell is responsible for finding bounding boxes.
- First download the pretrained weights and ncessary files from here If you are using linux, just type in the terminal
wget https://pjreddie.com/media/files/yolov3.weights
- Install necessary modules. Type in the terminal
pip install opencv-python
pip install numpy
pip install imutils
Now for webcam the default webcam port is 0. If you are using any external webcam, just change the port number with your ip address of ip camera or the external port.
First type this command in the terminal
ls /dev/video*
- Now eject the external webcam. Again type the same command.
ls /dev/video*
- You can see the a port is missing. That's the desired port of our external webcam.
Remember if you are using the builtin webcam of your laptop, you don't need to follow this process. After typing the command you will see only one port.
git clone https://github.com/goyeahia/object-detection-yolo
cd object-detection-yolo
Now replace all the downloaded weight to the "yolo" folder and type this command in the terminal
python object_detection.py
If you are having any os compatiblity issue, let me know. I will try to fix as soon as possible. This mini project has been successfully run on ArchLinux.