Stephen J. Kiselewich

Legendre and gabor moments for vehicle recognition in forward collision warning

Collision warning remains an active research field due to the increasing complexities of on-road traffic worldwide. Vision-based warning systems are of particular interest because of the extensive information contained in images. This paper proposes the combination of Legendre moments and Gabor features for monocular vision-based vehicle recognition. We focus on vehicle recognition within a region of interest (ROI) in an image by assuming that the ROI has been detected by a radar sensor. Two classifiers including a support vector machine (SVM) and a neural network have been investigated to verify the effectiveness of the features. We have tested the proposed approaches on real-world video sequences acquired under various weather conditions for a wide range of vehicles and non-vehicles at up to 70 meters. The proposed combination of Legendre moments and Gabor features has yielded a correct classification rate of 99.1% and a false alarm rate of 1.9%. We have compared the proposed features with the over-complete Haar wavelets in the literature

A monocular vision-based occupant classification approach for smart airbag deployment

Occupant classification is essential to a smart airbag system that can either turn off or deploy in a less harmful way according to the type of the occupants in the front seat. This paper presents a monocular vision-based occupant classification approach to classify the occupants into five categories including empty seats, adults in normal position, adults out of position, front-facing child/infant seats, and rear-facing infant seats. The proposed approach consists of image representation and pattern classification. The image representation step computes Haar wavelets 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 variety of indoor and outdoor images acquired under various illumination conditions 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.

Triangulation based technique for efficient stereo computation in infrared images

This paper proposes a technique to compute an accurate semi-dense disparity map from infrared stereo image pairs obtained using an uncalibrated stereo rig. First, an initial sparse disparity map is obtained using corner matching methods. This map is then refined using our proposed triangular constraints and the epipolar geometrical constraints to yield a more accurate semidense disparity map. Experimental results obtained using the proposed method are reported along with results obtained with a classical correlation based method as well as a more recent method based on graph-cuts. The proposed method yields good results even with low resolution, low texture infrared images. The proposed method is designed to be part of a vision-based occupant sensing system that will help to control airbag deployment in future vehicles.

Challenges of Embedded Computer Vision in Automotive Safety Systems

Vision-based automotive safety systems have received considerable attention over the past decade. Such systems have advantages compared to those based on other types of sensors such as radar, because of the availability of lowcost and high-resolution cameras and abundant information contained in video images. However, various technical challenges exist in such systems. One of the most prominent challenges lies in running sophisticated computer vision algorithms on low-cost embedded systems at frame rate. This chapter discusses these challenges through vehicle detection and classification in a collision warning system.