OpenCV多图像检测

Question

i'm trying to build a simple image detection algorithm in which will classify different objects such as plastic cups, bottles, cell phones, etc. What is the best way to implement this using openCV and python? Thanks

Answer 1

If you are looking for a algorithm or model that can detect various objects ie classify given image you have to look in to Conv Nets (CNN).

A CNN is a deep learning model that can be trained on the classes of images you want to detect. Later different images can be provided to the trained CNN to get the predictions.

If you are interested in this stuff, I recommend you to take a look in to this blog post.

You can try to do some pattern matching for the images but the accuracy will drastically reduce if there is heavy variations.

Answer 2

As far as I know you cannot train/build your own network on OpenCV. You could rather use OpenCV to load pre-trained models and perform detection on a set of images.

In order to create a new model you can use either caffe, torch or tensorflow frameworks and use it to perform detection in OpenCV.

In order to perform detection you need to do the following:

• First, ensure you have compiled OpenCV with the contrib module. If you don't simply enter the following onyour terminal: pip install opencv-contrib-python .
• First ensure you have cv2.dnn . If you don't you will need to update the version of OpenCV.
• Now you have an option to choose models from one of the following frameworks caffe, tensorflow, darkNet and torch:
  • readNetFromCaffe
  • readNetFromDarknet
  • readNetFromTensorflow
  • readNetFromTorch
• After choosing a framework select a pre-trained model. You can get pre-trained caffe models from HERE and tensorflow models from HERE . A good search on github will give you many more models.

I came across the dnn module after reading this blogpost by Adrian .

Just to re-iterate you can perform detection using pre-trained models on OpenCV but not create new models.

How do you run a pre-trained model?

1. caffe : model= cv2.dnn.readNetFromCaffe(ResNet.prototxt.txt, ResNet.caffemodel) , where the prototxt.txt file contains the architecture and the other file contains the model weights.

2. tensorflow : model = cv2.dnn.readNetFromCaffe(ResNet.pb, ResNet.pbtxt) , where .pb file containns the netwrok architecture and .pbtxt contains the model weights.

Refer to the link for more details.