Instance segmentation of real time video for object detection using hybrid Mask RCNN-SVM.

Detection of real-world factors in digital photos and videos is one of the most important challenges in computer recognition for object detection. The main goal of generic object detection is to identify and locate specific objects. Despite abundant benefits, there exist some problems, such as the accuracy and extraction of lower and higher-level features. In order to obtain the low and high level characteristics of an image for precise classification using Bi-directional Feature Pyramid Network (Bi-FPN) and Mask Regional based Convolutional Neural Network (Mask RCNN) with Support Vector Machine (SVM) are utilised. In this proposed model, the features are extracted with the Bi-FPN model and are pooled with the adaptive feature polling technique. These features are aligned with the ROI alignment and separated into convoluted and fully connected layers. The SVM replaces the fully connected layer for classification and the convoluted layer is used for masking and bounding the object with the box. This combined model with Mask RCNN and SVM represents the Hybrid Mask RCNN-SVM. The proposed model has been implemented in Python for calculating and comparing performance metrics such as the accuracy, error, precision and recall etc., for the proposed and existing model. The resultant values for accuracy, recall, error and precision for real-time object detection utilizing hybrid Mask RCNN-SVM are 0.98, 0.96, 0.027 and 0.98. Thus, the evaluation of performance metrics results that the values of the proposed model being better compared to the existing techniques. As a result, the proposed hybrid Mask RCNN-SVM model effectively segments the object from the real-time video. [ABSTRACT FROM AUTHOR]