Gait Based Human Identification in Bad Illumination

Vision-based human identification in bad illumination is one of the challenge. Gait based human identification is an important biometric feature. In order to study and analyze Gait recognition in bad illumination conditions a framework is proposed. Kinect sensor is used for signal extraction, which has the capability to track human in bad illumination. Database is created using 5 persons by considering good illumination conditions and bad illumination conditions. The features of each individual are extracted using skeleton information. Mean value of distance between centroid of left leg lower part and centroid of right leg lower part is proposed as a new feature for Gait recognition. Training and classification is performed using Levenberg-Marquardt back propagation algorithm and SVM algorithm. Mean value of distance between centroid of left leg lower part and centroid of right leg lower part provides good distinguishing values between different persons. Results obtained for both algorithms are tabulated. 82.66% recognition rate achieved with Levenberg-Marquardt back propagation algorithm where as 92% recognition rate achieved with SVM for 5 persons in bad illumination conditions, with fixed Kinect sensor set up. From results it is revealed that the proposed framework performs the gait based human identification in bad illumination.