Gabor/PCA/SVM-based face detection for driver’s monitoring

This article implements a face detection process as a preliminary step to monitor the state of drowsiness on vehicles drivers. We propose an algorithm for pre-detection based on image processing and machine learning methods. A Gabor filter bank is used for facial features extraction.
Nội dung trích xuất từ tài liệu:
Gabor/PCA/SVM-based face detection for driver’s monitoringJournal of Automation and Control Engineering, Vol. 1, No. 2, June 2013Gabor/PCA/SVM-Based Face Detection forDriver’s MonitoringDjamel Eddine Benrachou, Brahim Boulebtateche, and Salah BensaoulaUniversity Badji Mokhtar, Department of electronic, Annaba, Algeriadjamelben.univ@gmail.com, {bbouleb, bensaoula_salah}@yahoo.frAbstract—Driver fatigue cause each year a large number ofroad traffic accidents, this problem sparks the interest ofresearch to move towards development of systems forprevention of this phenomenon. This article implements aface detection process as a preliminary step to monitor thestate of drowsiness on vehicles drivers. We propose analgorithm for pre-detection based on image processing andmachine learning methods. A Gabor filter bank is used forfacial features extraction. The dimensionality of theresulting feature space is further reduced by PCA techniqueand then follows a classification of Face/No Face classesusing Support Vector Machine (SVM), for face detection.The simulation results on both databases namely PIE andORL datasets show the efficiency of this approach. Dimensionality reduction is adopted by PCA techniqueto create low dimensional features vectors for moreconvenient processing. SVM is used to extract relevantinformation from this low dimensional training data inorder to construct a robust specific classifier. This methodhas been tested on two available AT&T (ORL) and (PIE)Databases of human faces. The statistical evaluation ispresented for two different databases using both SVMskernels namely linear and Gaussian kernels, implementedseparately in order to detect the presence of a face or not.A. Proposed AlgorithmThe use of non-intrusive drowsiness detection methodsrequires several processing modules. In the proposedapproach a first step of extracting essential features of theface detection is performed by applying the Gaborsrepresentation on the image database. The advantage ofthis representation is that it allows us better spatialfrequency features localization. For the separation offeatures obtained by the Gabor filter bank, we use anSVM classifier.Dimensionality reduction is applied using PCA tocreate low dimensional features vectors for moreconvenient processing. For the separation of the reducedfeatures obtained by Gabor filter bank, we use an SVMclassifier. The classification will be followed by adynamic neural network module (TDNN). This phase ofdecision takes into account the dynamics of yawning andblinking. The structure of this algorithm is illustrated in“Fig. 1”.Index Terms— drowsiness, car driver, face detection, gaborfilter, PCA, SVM classifierI.INTRODUCTIONDrivers drowsiness causes each year a large number ofroad traffic accidents. Statistics show that 10% to 20 % ofaccidents overall road traffic are due to the decrease levelof drivers alertness [1].The hypovigilance reduce the capacity to react, judgeand analyze information and it is often caused by fatigueand/or drowsiness. However fatigue and drowsiness aredifferent. The first one refers to a cumulative processproducing difficulty to pay attention while the second oneconcerns the inability to stay awake. Therefore, it isimportant to monitor drivers vigilance level and issue analarm when he is not paying attention.Monitoring drivers responses are approached by a lotof methods, sensing physiological characteristics, driveroperations, or vehicle response. These methods work welland give good indicators of vigilance state.Recently, we can find detection systems using vehicleembedded cameras [2]. These systems analyze visualcues generated by the drowsiness such as eye blinking,the drivers gaze or positioning of the drivers head [3]because the decline of the head could be a good indicatorof drowsiness.This paper investigates the ability of Gaborrepresentation and Support Vector Machine for visualfeatures extraction and captures the importantinformation by discriminatory method for face detectiontask. The idea is to decompose a face image into differentspatial frequencies (scales) and orientations where salientdiscriminant features may appear.1Figure 1. Flowchart of the detection algorithm (detecting the state ofdriver’s drowsiness by analysis of visual cues).Manuscript received October 15, 2012; revised December 22, 2012.©2013 Engineering and Technology Publishingdoi: 10.12720/joace.1.2.115-118115Journal of Automation and Control Engineering, Vol. 1, No. 2, June 2013B. Feature ExtractionThe features extraction step consists in transformingthe input raw data into meaningful information. Thus, weobtain a reduction of the decision space which mayaccelerate the processing time. The Gabor filters providea simultaneous representation in spatial and frequencydomain. This representation is an optimal tool used forthe purpose of local features extraction. It is efficientbecause it produces the same operating principle ofsimple cells in visual cortex of the mammal’s brain andalso properties in multi-directions, optimal for measuringlocal spatial frequencies. Based on these advantages, theGabor representation is widely used in applications ofimage analysis and applications of face recognition [4], [5]and extraction of features such as facial expressions.The family of complex Gabor wavelets could berepresented as follows:2  , ( x ) k  ,2exp 2k  ,2x22  2   exp  ik  , .x   exp    2  (1)where  is the standard deviation of Gaussian kernel, and  define the orientation and scale of Gabor filterD. Support vector MachineSupport vector machines (SVMs) are a very popularmethod for binary classification. The support vectorclassifier chooses one particular solution, the classifierwhich separates the c ...