Countour based HOG deer detection in thermal images for traffic safety

Car accidents due to deer vehicle crashes (DVCs) are constantly a major safety issue for the driving on rural roads in the Europe and North America. Many attempts have been made to avoid these accidents, but few have been succeeded. One of the options is to use thermal images to identify the presence of deer. In this approach, large amount of image data have to be processed. This process will take long time and make the system incapable of real-time deer identification and tracking. Based on the well-known pattern recognition methods - histogram of orientated gradient (HOG) method and support vector machine method (SVM), this research developed a contour based HOG + SVM method, called CNT-HOG method. Experimental results showed that this algorithm has the advantages of both high accuracy (up to 95%) and less consumed time (0.1s) than traditional HOG+SVM method. Further analysis has been carried out to evaluate the influence of body postures and body blockages. By achieving this speed and accuracy, deer can be tracked in real-time. Accurate tracking and identification can then be achieved to reduce the possibility of deer - vehicle crashes.