Human Figure Detection based on Convolutional Neural Network (CNN) and Support Vector Machine (SVM)

In many computer vision applications, human figure detection is one of the crucial tasks to implement due to its high functionality in a real-life application, such as surveillance systems, human-computer interaction, and autonomous vehicles. However, accurately distinguishing human from non-human objects in input images remains a challenging task, primarily attributed to the complexity and variability of visual data. The system must be capable of achieving high performance in terms of accuracy and processing time to be applicable in real-time scenarios. Addressing this formidable challenge becomes the focal point of the research, where developing a robust and efficient system is strive to accurately classify input images and extract discriminative features in order to distinguish between human and non-human objects. In this study, convolutional neural networks (CNN) and support vector machines (SVM), two well-liked methods for video segmentation, are thoroughly reviewed in this paper, where a system trained from both of these classifiers were done to create a detection system for human and non-human, along with analyzing their performance. At the end of the project, CNN method achieves an accuracy of 99.07%, processing time per frame of 253.25ms and f-score of 0.9908 while SVM-HOG achieves accuracy of 86.79%, processing time per frame of 0.54ms and a f-score of 0.8724.