The Research Of Nighttime Pedestrian Detection Technology

Due to the poor light condition, night-time driving is more dangerous than day-time driving. And there are worrisome leaks of security monitoring at night. To improve the safety of night driving and the accuracy of the intelligent monitoring at night, a growing number of studies begin to focus on night-time pedestrian detection technology. Pedestrian detection technology based on infrared images has a great potential because the discriminability of pedestrian against background in infrared images is more outstanding than visible spectrum images. Nevertheless, pedestrian detection is difficult because of poor infrared image quality (low resolution, low contrast, few distinguishable feature points, little texture information, etc.) and misleading signals.The histograms of gradient orientations (HOG) and the histograms of local intensity differences (HLID) are good single features for pedestrian detection, but it’s difficult to distinguish pedestrian from object whose contour is similar to pedestrian. Furthermore, the HOG and HLID do not take good advantage of the brightness information that inside the person’s contour in infrared images. To address these problems, based on the study of some common features in pedestrian detection now, this paper put forward a new feature for pedestrian detection in infrared images. With the help of support vector machine (SVM), the feature of histograms of oriented gradients and intensity (HOGI) is proposed through combining the best of HOG and histograms of intensity (HOI).The HOGI is a new feature which is very suitable for pedestrian detection in infrared images. The experiments show that when the false alarm rate is same the miss rate of HOGI is reduced by50%than HOG without any increase in feature dimension, and the HOGI increases the computation efficiency by an average of40%.Considering the real-time requirement in applications such as the field of intelligent vehicle driving, and combining the laser radar which is installed in the vehicle, this paper further proposed a pedestrian detection method which is based on the fusion of laser radar and infrared images. In the system, firstly, extracting obstacles in the radar data, and getting the regions of interest through the preliminary pedestrian recognition by using geometric features. The regions of interest in radar are projected to infrared images, so the regions of interest in infrared images are acquired. Then start pedestrian detection in these regions of infrared images with the HOGI feature. This system has further improved the real-time property and accuracy of pedestrian detection.