In the field of computer vision, object detection is getting more attention due to its huge applications in visual monitoring. Multiple object detection identifies the position of objects or regions of objects in the image or videos. Many methods were developed for detecting multiple objects, but the overall detection accuracy of those methods was limited due to the congested environment, complex background, and similarities between the objects. To solve such an issue, this research study proposed the feature extraction method for multiple object detection using Histogram of Oriented Gradient (HOG) with Local Ternary Pattern (LTP). The Caltech 101 dataset is used in the proposed method where the images are converted to LAB. The process of feature extraction takes place by using the proposed HoG and LTP to detect prominent regions from the image. Further, the obtained features are fused by using Deep Convolutional Neural Network (D-CNN) and then forwarded to Region-based Convolutional Neural Network (R-CNN) to detect the multiple objects. The proposed HoG and LTP feature extraction method has the advantages of improving the classification accuracy by effectively extracting the oriented features and texture features. The proposed method achieved better accuracy of 92.48%, whereas the existing Multi-Object Detection and Tracking (MODT) method achieved an accuracy of 76.23% for the detection of multiple objects.

No abstract

No abstract

No abstract