Face and vehicle recognition

Question

I am looking for a library (open source or not) for analyzing photographs and determining the presence of vehicles and people in the images. I am not looking for face recognition or car classification. Just the ability to determine with a certain degree of certainty that the vehicle and / or person is contained in images taken periodically.

I can provide a significant number (1000) of photos for training the system.

Change 1

I expect to send photos to the training system and understand that the quality of the detection system will depend on my training photos. I think this is what I see in the Numenta Image demo software (well packed), but with a more flexible api (learning and calling the system from the http api would be nice).

Are there any implementations of the OpenCV haar system that have been well complemented by the Windows program?

Answer 1

OpenCV is definitely one of the best options you can start with.

But , pay attention that what you want to achieve is really far from what you think. It doesn't look like a library, say, JPEG compression, when you need a day to understand and use. OpenCV is a set of tools and algorithms for scientists and engineers to research new ways of image recognition; You may need a significant amount of time and energy to complete your task.

Answer 2

Use search terms such as machine learning, neural network, image classification, library or package, and your favorite programming language. Maybe anything from scikit-learn to Weka.

Also see here Image Classification Algorithms Using Java

Answer 3

OpenCV can provide you with training materials in Haar cascades for discovery. Here is a tutorial for haartraining. Please note that this is an old haartraining program, not a new traincascade program (it supports multithreading, a quick search should show you how to use it, it's almost the same use).

Answer 4

The first approach, similar to human detection, is to study the classifier of machine linear support machines on a histogram of oriented gradient descriptors. Here is a basic article from INRIA . Opencv provides an implementation of HoG, linear SVM, and an evaluation framework.