THE WAY OF REAL-TIME PEDESTRIAN RECOGNITION AND TRACKING USING KALMAN FILTER AND CLUSTERING ALGORITHM BASED ON CASCADE METHOD BY HOG

摘要

A pedestrian recognition technique relates to a technique for searching for a pedestrian in an image and is widely applied for common security and monitoring such as alarming a risk by recognizing a person in the front, counting the number of persons passing by one point, keeping the safety of workers in a workspace, etc. A basic method in a current pedestrian detection technique is a histogram of oriented gradients (HOG) which maintains the best performance. However, a HOG algorithm has a disadvantage of being hardly applied in real time and being slow due to a large quantity of operation. The present invention relates to the pedestrian recognition technique through a camera which is installed on a vehicle. The proposed method implements a cascade HOG algorithm based on data previously learned in an interest area. A window suitable for the pedestrian size is detected as a Haar-like algorithm for a window area which is detected to the HOG cascade for resizing of the window area. The average value of the window size detected from the cascade HOG and detected through the Haar-like is determined as a pedestrian window at an optimum level of the window size for solving a problem of imaging only the upper body or the lower body. Each pedestrian window is traced by applying a Kalman filter in the previous algorithm, and an overlapped area of the detected pedestrian is removed by applying a clustering method on a mutually closed window so as to increase the recognition ratio of the pedestrian.