Limits...
Optical flow and driver's kinematics analysis for state of alert sensing.

Jiménez-Pinto J, Torres-Torriti M - Sensors (Basel) (2013)

Bottom Line: Our experiments show the effectiveness of the approach with a correct eyes detection rate of 99.41%, on average.The results obtained with the proposed approach in an experiment involving fifteen persons under different levels of sleep deprivation also confirm the discriminability of the fatigue levels.In addition to the measurement of fatigue and drowsiness, the pose tracking capability of the proposed approach has potential applications in distraction assessment and alerting of machine operators.

View Article: PubMed Central - PubMed

Affiliation: Department of Electrical Engineering, Pontificia Universidad Catòlica de Chile, Vicuña Mackenna 4860, Casilla 306-22, Santiago, Chile. jejimenp@puc.cl

ABSTRACT
Road accident statistics from different countries show that a significant number of accidents occur due to driver's fatigue and lack of awareness to traffic conditions. In particular, about 60% of the accidents in which long haul truck and bus drivers are involved are attributed to drowsiness and fatigue. It is thus fundamental to improve non-invasive systems for sensing a driver's state of alert. One of the main challenges to correctly resolve the state of alert is measuring the percentage of eyelid closure over time (PERCLOS), despite the driver's head and body movements. In this paper, we propose a technique that involves optical flow and driver's kinematics analysis to improve the robustness of the driver's alert state measurement under pose changes using a single camera with near-infrared illumination. The proposed approach infers and keeps track of the driver's pose in 3D space in order to ensure that eyes can be located correctly, even after periods of partial occlusion, for example, when the driver stares away from the camera. Our experiments show the effectiveness of the approach with a correct eyes detection rate of 99.41%, on average. The results obtained with the proposed approach in an experiment involving fifteen persons under different levels of sleep deprivation also confirm the discriminability of the fatigue levels. In addition to the measurement of fatigue and drowsiness, the pose tracking capability of the proposed approach has potential applications in distraction assessment and alerting of machine operators.

Show MeSH

Related in: MedlinePlus

Example of LK tracking between two frames, showing an incorrectly tracked point.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3673081&req=5

f6-sensors-13-04225: Example of LK tracking between two frames, showing an incorrectly tracked point.

Mentions: The driver's pose estimation problem consists in finding the pose angles α1, β1, γ1, α2 and β2 at time instant k + 1, given the knowledge of the pose at time k and the driver's motion Mk+1 at instant k + 1 as measured from the image. The proposed approach to estimate and track the driver's pose angles employs the Lucas-Kanade's (LK) method to optical flow computation [57]. The LK method computes a set Sk of salient points, pj,k, j = 1,2,…,N, in an image frame at instant k and tracks point-by-point yielding weights Wj,k+1 and a set Sk+1 of salient points pj,k+1, j = 1, 2, …,N, in the image frame at instant k + 1 corresponding to the points pj,k, j = 1, 2,…,N, in the previous frame, as illustrated in Figure 6. The velocity at which corresponding pixels move from pj,k to Pj,k+1, approximated by ϕj,k+1 = (Pj,k+1 − pj,k)/Ts for a sampling period Ts, is the so-called optical flow of the image's intensity at pixel pj,k at time instant k + 1. Each weight Wj,k+1, j = 1, 2,…,N, is a measure of the similarity between a pair of corresponding points pj,k and pj,k+1 computed as the convolution of pixel neighborhoods surrounding pj,k and Pj,k+1. The weights Wj,k+1, j = 1, 2,…, N, provide a measure of the quality and reliability of the match and are particularly useful to discard points with lower weights, which are more likely to occur near the boundaries of the SPG when the head turns, as some points will become occluded. Bad tracking of some of the salient points, as depicted in Figure 6, may also occur when the points are occluded by an external object, like the driver's hand, or when weak saliency, due to low textureness or contrast, makes correspondences ambiguous (non-unique).


Optical flow and driver's kinematics analysis for state of alert sensing.

Jiménez-Pinto J, Torres-Torriti M - Sensors (Basel) (2013)

Example of LK tracking between two frames, showing an incorrectly tracked point.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3673081&req=5

f6-sensors-13-04225: Example of LK tracking between two frames, showing an incorrectly tracked point.
Mentions: The driver's pose estimation problem consists in finding the pose angles α1, β1, γ1, α2 and β2 at time instant k + 1, given the knowledge of the pose at time k and the driver's motion Mk+1 at instant k + 1 as measured from the image. The proposed approach to estimate and track the driver's pose angles employs the Lucas-Kanade's (LK) method to optical flow computation [57]. The LK method computes a set Sk of salient points, pj,k, j = 1,2,…,N, in an image frame at instant k and tracks point-by-point yielding weights Wj,k+1 and a set Sk+1 of salient points pj,k+1, j = 1, 2, …,N, in the image frame at instant k + 1 corresponding to the points pj,k, j = 1, 2,…,N, in the previous frame, as illustrated in Figure 6. The velocity at which corresponding pixels move from pj,k to Pj,k+1, approximated by ϕj,k+1 = (Pj,k+1 − pj,k)/Ts for a sampling period Ts, is the so-called optical flow of the image's intensity at pixel pj,k at time instant k + 1. Each weight Wj,k+1, j = 1, 2,…,N, is a measure of the similarity between a pair of corresponding points pj,k and pj,k+1 computed as the convolution of pixel neighborhoods surrounding pj,k and Pj,k+1. The weights Wj,k+1, j = 1, 2,…, N, provide a measure of the quality and reliability of the match and are particularly useful to discard points with lower weights, which are more likely to occur near the boundaries of the SPG when the head turns, as some points will become occluded. Bad tracking of some of the salient points, as depicted in Figure 6, may also occur when the points are occluded by an external object, like the driver's hand, or when weak saliency, due to low textureness or contrast, makes correspondences ambiguous (non-unique).

Bottom Line: Our experiments show the effectiveness of the approach with a correct eyes detection rate of 99.41%, on average.The results obtained with the proposed approach in an experiment involving fifteen persons under different levels of sleep deprivation also confirm the discriminability of the fatigue levels.In addition to the measurement of fatigue and drowsiness, the pose tracking capability of the proposed approach has potential applications in distraction assessment and alerting of machine operators.

View Article: PubMed Central - PubMed

Affiliation: Department of Electrical Engineering, Pontificia Universidad Catòlica de Chile, Vicuña Mackenna 4860, Casilla 306-22, Santiago, Chile. jejimenp@puc.cl

ABSTRACT
Road accident statistics from different countries show that a significant number of accidents occur due to driver's fatigue and lack of awareness to traffic conditions. In particular, about 60% of the accidents in which long haul truck and bus drivers are involved are attributed to drowsiness and fatigue. It is thus fundamental to improve non-invasive systems for sensing a driver's state of alert. One of the main challenges to correctly resolve the state of alert is measuring the percentage of eyelid closure over time (PERCLOS), despite the driver's head and body movements. In this paper, we propose a technique that involves optical flow and driver's kinematics analysis to improve the robustness of the driver's alert state measurement under pose changes using a single camera with near-infrared illumination. The proposed approach infers and keeps track of the driver's pose in 3D space in order to ensure that eyes can be located correctly, even after periods of partial occlusion, for example, when the driver stares away from the camera. Our experiments show the effectiveness of the approach with a correct eyes detection rate of 99.41%, on average. The results obtained with the proposed approach in an experiment involving fifteen persons under different levels of sleep deprivation also confirm the discriminability of the fatigue levels. In addition to the measurement of fatigue and drowsiness, the pose tracking capability of the proposed approach has potential applications in distraction assessment and alerting of machine operators.

Show MeSH
Related in: MedlinePlus