deer detection 1
Computer Vision, OpenCV and HaarTrainng
As the starting point of this vehicle safety system, animal detection is an essential and indispensable technology. Other implementations, head-up display and led lighting, will be responded by this detecting sensor. For detecting animals, most efficient device is computer vision that we usually call it as a webcam. Computer vision could be applicable to object recognition, identification, detection, content-based image retrieval, pose estimation optical character recognition, tracking, optical flow, scene reconstruction, image restoration, 3D volume recognition and etc. Object recognition is the technology that will be used as animal detection. And there is OpenCV that is a library of programming functions mainly aimed at real time computer vision. OpenCV was developed by intel but now it’s free with BSD license.
OpenCV provides a statistical machine-learning library. One of the most popular libraries is face detection, one of haar-like feature. It is a trained haarcascade library that is converted into haarcascade_frontalface_alt.xml file. OpenCV also provides HaarTraining application. By this training, we can make own trained haarcascade xml files that will be the object recognition libraries. To make own haar trained xml file is seriously complicate and has multiple steps, also it is needed lots of pictures. For the face detection, we usually use, the researcher used a total of 10,000 pictures, 5000 positive pictures and 5000 negative pictures. Even OpenCv doesn’t provide specific and comprehensible tutorial about HaarTraining, and everywhere I searched on web says the usual processing period is 6 days. I know if I you thousands images, it will take long. However, this project might be used with night vision. It means animals would be presented closely as silhouette, not serious inside detail. This research is summery of the complicate and difficult processing of HaarTraining.
