Pedestrian detection by video processing in automotive night vision system

The paper takes the automatic detection of pedestrians from a vehicle into consideration using an automotive night vision system. Accidents involving pedestrians at night cause a significant part of deaths in the roads. Therefore, the authors present a design of the automotive night vision system that combines the passive solutions, i.e. those using thermal vision, and active, which use the near infrared camera. Passive system provides greater range of detection, while the active one works in a closer range and offers more natural images. Beside the design of hardware also a software part for the automatic detection of pedestrians was prepared and tested. The software for object detection and classification uses modern digital signal processing algorithms like: connected component labeling (CCL), histogram of oriented gradients (HOG), and support vector machine (SVM). Tests performed on real night vision recordings show a very high efficiency of the proposed solution thus confirm correctness of the design. Furthermore, the presented night vision system for automatic detection of pedestrians can be implemented in vehicles at an acceptable, i.e., quite moderate cost.