A novel visible spectrum images-based pedestrian detection and tracking system for surveillance in non-controlled environments

For many vision applications, robust detection and tracking of pedestrians in image sequences are essential. In this paper, a hybrid system for pedestrian detection and tracking is presented. The proposed method is achieved in four major stages. First, the given image is segmented by exploiting the motion information to generate regions of interests where pedestrians are likely to exist. In the second stage, the ROIs are subjected to a selection process in order to keep only significant ones. Then, a hybrid feature is defined to classify the ROIs generated from the previous step. It entails combining the so-called GLBP-Color together with the histograms of oriented optical flow (HOOF). Artificial neural networks (ANN), Adaboost and support vector machine (SVM) classifiers have been tested together with the introduced hybrid feature and the first one is adopted as it outperforms the other two. The last step performs the tracking of the detected pedestrians using the so-called Color-BMA, which is an extension of the classical block matching (BMA) algorithm to the RGB color space. The proposed method has been tested in non-controlled environments with a collection of common databasets that are well known in the surveillance research community (CAVIAR, PETS 2006 and PETS 2009). The obtained results are satisfactory when compared to the recent state of the art approaches.

Automated summary

This paper presents a hybrid system for robust detection and tracking of pedestrians in image sequences. The proposed method utilizes motion information to segment the given image and generate regions of interests (ROIs) where pedestrians are likely to exist. A hybrid feature combining GLBP-Color and histograms of oriented optical flow (HOOF) is defined to classify the ROIs. The tracking of detected pedestrians is performed using a modified block matching algorithm in the RGB color space, called Color-BMA. Experimental results on common databases show that the proposed method outperforms the state-of-the-art approaches.