No abstract

Automatic detection of moving objects at distance and in all weather conditions is a critical task in many visionbased safety applications such as video surveillance and vehicle forewarn collision warning. In such applications, prior knowledge about the object class (vehicle, pedestrian, tree, etc.) and imaging conditions (shadow, depth) is unavailable. What makes the task even more challenging is when the camera is non-stationary, e.g., mounted on a moving vehicle. The essential problem in this case lies in distinguishing between camera-induced motion and independent motion. This paper proposes a robust algorithm for automatic moving object detection at distance. The camera is mounted on a moving vehicle and operates in both day and night time. Through the utilization of the focus of expansion (FOE) and its associated residual map, the proposed method is able to detect and separate independently moving objects (IMOs) from the "moving" background caused by the camera motion. Experimentations on numerous realworld driving videos have shown the effectiveness of the proposed technique. Moving objects such as pedestrians and vehicles up to 40 meters away from the camera have been reliably detected at 10 frames per second on a 1.8GHz PC.

Frontal "depowered" air bag systems are underdesign today to be even more effective than current air bags in saving lives, while at the same time reducing the potential of causing an air bag induced serious injury or death. Stereovision real-time occupant sensing systems (airbag suppression) have been developed at Delphi Automotive Systems 1 for use in comercial vehicle applications. One of the issues in such a system is that the irrelevant non-stationary background information within the field of view of the cameras results in less robust occupant classifications. In this paper, a disparity-based image segmentation method is provided. The input images are first segmented according to a pre-determined disparity threshold map and the real-time disparities of the occupants. Binary image processing techniques are used to reject noise introduced into the segmented images through low-resolution disparity calculations. The segmented images are then used for image feature extraction for a neural network classifier. For comparison, two neural network classifiers were created with and without the infrared image segmentation. Our experiements on segemnted images shown an increase of the classifier performance by at least 23% on a large database of IR images collected in extreme outdoor conditions.

Occupant classifcation is essential to a smart airbag system that can either turn off or deploy in a less harmful way according to the p p e of the occupants in the front seat. This paper presents U monocular vision-based occupant classification approach to d a s s i h the occupants into five categories including empty seats, adiilts in normal position, adults out of position, front-facing chila7infant seats, and rear-facing infant seats. The proposed approach consists oj image representation and pattern classification. The image representation step computes Haar wavelers and edge features from the monochrome video frames. A support vector machine (SVM) classifier next determines the occupant category based on the representative features. We have tested our approach on a large variew of indoor and outdoor images acquired under various illumination cotadifions for occupants with different appearances, sizes and shapes. With a strict occupant exclusive training/testing split, our approach has achieved an average correct classification rate of 97.18% among the five occupant categories.

No abstract