DEVICE AND METHOD FOR TRACKING PEDESTRIAN IN THERMAL IMAGE USING AN ONLINE RANDOM FERN LEARNING

摘要

The present invention relates to a device and method using the online random fern learning in thermal imaging to track the pedestrian, is more specifically to a pedestrian in the thermal image input from the camera, obscured pedestrian shape change or load occurs or similar exists to be separated from the background and other pedestrian to the pedestrian tracking, to an apparatus and method for tracking a pedestrian by using the online learning random fern in thermal imaging. According to the apparatus and method in a thermal imaging proposed in the present invention by using the online random fern study to track the pedestrian, the thermal image from the image input from the camera Haar-like feature based random forest with OCS-LBP the feature based Random Forest used to detect the pedestrians and, using the detected pedestrian OCS-LBP features and characteristics LID online random fern learning and particle filters based on, it can be tracked in real time the number of pedestrians. Further, according to the present invention, through the particle filter iteratively estimating the subsequent location distribution of the pedestrians, can be more easily in pedestrian tracking, also, to measure the observation likelihood for a particle weight is determined, the normal distance measuring device instead of using the online learning random fern, it can increase the success rate tracking. Furthermore, based on the initial for the fern perform the online learning, and next, the position distribution of the random fern this, the pedestrian of the selected fern in subsequent frames observed for tracer model using, Boosted random fern, according to the invention can solve that by reducing the amount of sample required a time-consuming learning by making learning materials. In addition, According to the present invention, through the distance between the detected pedestrian and tracker, a relevance check, based on a calculated value of a combination of odds and overlap ratio of the detected pedestrian by using the model of the tracer, maintaining the identity of the pedestrian, tracking the success rate can be improved. Furthermore, according to the present invention, based on the pedestrian detection application associativity line scan algorithms and random fern study, the overlap of a pedestrian or a moving camera is excellent in tracking performance for the pedestrian environment, as well as newly appearing. ;