Prediction of Human Misbehavior Detection From CCTV Video Recording Using Convolutional Neural Network Compared With VGG-16 Method
N. Jhansi, Shaik Mahaboob Basha · 2023
In contrast to the Visual Geometry Group (VGG-16) technique, this study uses a novel implementation of Novel Convolutional Neural Network (CNN) algorithm to increase the accuracy of human misbehavior detection in real-time Closed-Circuit Television(CCTV) footage. Materials and Methods: A total of 3200 samples used for this research is obtained from Kaggle database. The Kaggle database system served as the source for the research dataset used in this study.The study utilized G-power analysis with 80% power,alpha rate of 0.05 results in 20 iterations for each group with total iterations as 40. The clincalc tool was used to determine the total number of iterations N equals 40 (20 iterations for each group). In G-power analysis the power utilized was 80%, alpha rate was 0.05 and the sample size was 3200(for 2 groups). CNN and VGG-16, both with the same amount of data samples (N=10), are used to perform the prediction of an individual's misbehavior, with CNN achieving a better accuracy rate. Results: The proposed CNN achieves a 97.25 percent success rate compared to the VGG-16 classifier's 90.30 percent success rate. This represents a significant difference. The significance level of the investigation was determined to be p=0.001(p<0.05) which is significant statistically. Conclusion: Suggested CNN model significantly obtains a greater rate of accuracy than the VGG-16 model when it comes to the prediction of human misbehavior in real-time surveillance video.