Emotion recognition of the driver based on KLT algorithm and ShuffleNet V2

Emotion recognition of the driver based on KLT algorithm and ShuffleNet V2

연구 분야: Verification

학회: Signal, Image and Video Processing

Emotional monitoring was essential in the development of sophisticated automobiles with advanced driver assistance systems (ADAS) to ensure safety and monitor potential collision trends for evaluating the driver's mental state. Factors affecting driver emotional identification include posture changes, illumination, and occlusions. Existing emotion recognition using CNN ResNet101 has low sensitivity, high false-positive rate, and error. To overcome these challenges, this paper proposes Driver Emotion Recognition (DER) using ShuffleNet V2 to effectively recognize emotions and determine the driver's mental condition. Initially, the facial images from different peoples are collected as a dataset and pre-processed using image resizing, Gaussian filter, median filter, histogram equalization and wiener filter for removing noise and enhancing the image quality. For segmentation and feature extraction of face images from a variety of datasets such as CK_Plus, FER_2013, TFEID, KMU_FED, and KDEF, the Region of Interest (ROI) and Kanade-Lucas-Tomasi (KLT) algorithm are used, which segments the face images based on region. Then, the ShuffleNet V2 classification is used to categorize emotions into six unique expressions such as happy, surprise, sad, fear, anger, disgust, and neutral. The performance of the proposed model is assessed by comparing it to that existing models. The proposed approach achieved an accuracy rate of 0.98% in CK_Plus, 0.97% in FER_2013, 0.97% in TFEID, 0.99% in KMU_FED, and 0.99% in KDEF. In comparison to other existing techniques, the proposed technique performs better. In order to determine the different emotions, the created model is the best choice.