Phase-edge based approach for pedestrian segmentation using NIR camera and tracking for driver assistance

A new approach using phase edge based techniquefor night time pedestrian segmentation and tracking using Near IR camera is proposed in this paper. A Gabor filter is used to extract even weak potential vertical edges that belong to candidate pedestrian blocks even under non-uniform illuminations and poor contrast. Vertical and horizontal projection on Gabor filtered response is carried out to determine the bounding-box of pedestrian-like candidate blobs. Pedestrian like candidate blocks undergo through a series of rule based classifiers. Once such set of pedestrian blocks are identified, candidate pedestrian blocks are tracked in next consecutive frames of a video using density confidence based tracker which uses Gabor filter edge response for feature-space analysis. False positives are greatly removed using spatio-temporal analysis and tracking mechanism. Performance characterization of the algorithm has been carried out in highway roads under nonuniform illuminations and varying contrast. Experimental result shows that algorithm can produce a detection rate of nearly 69% and can be used in real-time.