Abstract
Traffic accidents make car safety receive most attention in recent years. With the progress of image processing technology, the automotive safety equipment sets cameras on cars and conducts image processing with the images captured by the cameras, which provides drivers more traffic information. In the image-based active driving safety equipment, the pedestrian detection technology is important. Most previous works that used cameras to capture traffic images applied classifiers to train the pedestrian features and conduct multi-stage feature matching. In our proposed method, we apply the single-lens camera to capture images, and we use formulae as well as image processing to extract the pedestrian features of each body part, such as edge line detection and color grouping. As a result, we exclude the objects that are not pedestrians on the road and find the correct pedestrians. Regarding the performance, the proposed method saves the computation time for manual template selection and pedestrian feature training of classifiers, which meets the requirements of real-time processing. The proposed method also provides the benefits to change cameras without conducting the above procedure repeatedly and only adjusts according to the pedestrian size. The result shows that the average computation time of pedestrian detection speed of the proposed method achieves 82.43 fps on Intel Core i7 processor at 3.4 GHz, the detection rate is better than 88%, and the false positive rate is no more than 10%.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have