Intensity self similarity features for pedestrian detection in Far-Infrared images

Abstract

Pedestrian detection is an important but challenging component of an Intelligent Transportation System. In this paper, we describe a pedestrian detection system based on a monocular vision with a Far-Infrared camera (FIR). We propose an original feature representation, called Intensity Self Similarity (ISS) , adapted to pedestrian detection in FIR images. The ISS representation is based on the relative intensity self similarity within a pedestrian region of interest (ROI) hypothesis. Our system consists of two components. The first component generates pedestrian ROI hypothesis by exploiting the specific characteristics of FIR images, where pedestrian shapes may vary in large scale, but heads appear usually as light regions. Pedestrian ROI are detected, with high recall rate, due to a Hierarchical Codebook (HC) of Speeded-Up Robust Features (SURF) located in light head regions. The second component consists of pedestrian hypothesis validation, by using a pedestrian full-body classification based on the ISS representation, with Support Vector Machine (SVM). For classification, we retained two feature descriptors: the Histogram of Oriented Gradients (HOG) descriptor and the original ISS feature representation that we proposed for FIR images. The early fusion of these two features enhances significantly the system precision, attaining an F-measure for the pedestrian class of 97.7%. Moreover, this feature fusion outperforms the state-of-the-art SURF descriptor proposed previously. The experimental evaluation shows that our pedestrian detector is also robust, since it performs well in detecting pedestrians even in large scale and crowded real-world scenes.

SWIR images evaluation for pedestrian detection in clear visibility conditions

“…The most frequently used are visible cameras either in a monocular ( [1], [5], [6]) or a stereo configuration ( [7], [8], [2]). Infrared cameras (mainly working in the thermal spectrum) have too been successfully employed for pedestrian detection systems ( [9], [10], [11], [12], [13]). An extended review of different pedestrian detection systems for ADAS based on a variety of sensors is given in [14].…”

Far-infrared pedestrian detection for advanced driver assistance systems using scene context

“…A considerable amount of previous work has been carried out in recent years for pedestrian detection using a monocular vehicle-mounted FIR camera. (3) In the classification phase, most FIR pedestrian classifiers are obtained through a machine learning algorithm which is based on labeled training samples of pedestrians and nonpedestrians, such as support vector machine (SVM), 7,8 AdaBoost, 15 and random forests. Specifically, (1) in the ROI generation phase, major methods include template matching, 14 intensity segmentation, 15,16 and intensity-oriented projection.…”

“…Recently, various researchers have considered the use of low-level features for FIR pedestrian detection, such as pyramid binary pattern (PBP), 2 pyramid entropy weighted histograms of gradients (PEWHOG), 3 intensity self similarity (ISS), 7 histogram of oriented phase energy (HOPE), 8 and so on. They only use low-level features to depict appearance patterns of pedestrians for performing pedestrian detection.…”

Pedestrian detection in infrared images using HOG, LBP, gradient magnitude and intensity feature channels

“…Their work is extended by [21] that add several fast computing global features that improve the discriminative power representation. Intensity Self Similarity (ISS), adapted to pedestrian detection in far IR images is proposed by [22]. The ISS encodes the distribution of color as repetition across the image.…”