Object-Oriented Scale-Adaptive Filtering for human detection from stereo images

Abstract

In this paper, an effective and efficient methodology to extract visual evidence of suitable scale for object detection, Object-Orient Scale-Adaptive Filtering (OOSAF), is proposed. With OOSAF, object extraction from stereo images is formulated as the design of scale-adaptive filters. Based on OOSAF, two methods for human detection from stereo images are developed. One is to detect human objects with close distances to the camera for intelligent human-machine interaction, and the other is to detect human heads in distant crowds for security surveillance. Experiments show that, with OOSAF, efficient solutions for human detection from stereo images could be achieved with high detection rates and low false alarm rates.