A CNN-RF Hybrid Model for Cross View Gait Recognition Method using CASIA-B Cross

Many advancements in human Gait have been made so far in the research field which have even been implemented to a greater extent. Gait recognition has proved to be much more compatible as compared to other biometric processes. Currently, cross-view gait identification techniques disregard the possible dynamical information of the gait sequences and always create representations for recognition from different deep convolutional networks. If pedestrians possess distinct gait patterns, gait recognition may be executed by the computation of their dynamic properties from every perspective. The problem of cross-view gait detection is difficult because variations in perspective can significantly affect the silhouettes of gaits. Deep learning advancements have raised the bar for cross-view gait identification performance. However, the scarcity of gait samples from various viewpoints limits the effectiveness of the deep learning-based cross-view gait detection techniques now in use. This paper uses the methodology provided in Deep Learning algorithms through CNN and Random Forest. After the role played by different layers in CNN, a strong classification is performed by Random forests by detecting a strong probability based on more than one decision tree. CASIA B Gait dataset has been used that included 6703 subjects, from four view angles CV1, CV2, CV3, and CV4 $\left(15^{\circ}, 30^{\circ}, 45^{\circ}, 60^{\circ}\right)$ Here the model has been used to classify gait silhouettes at Four view angles CV1, CV2, CV3, and CV4. A performance metrics have been generated as an output of the classification of the subjects. Overall Cross View Angle 2 CV2 had the maximum accuracy out of all the four view angles which is $0.98 \%$, the second highest accuracy was CV1 i.e. $0.97 \%$, CV3, and CV4 got the same value of accuracy i.e. $0.96 \%$. Furthermore, we assess several gait recognition methods that are resistant to view angle variations.