Title: Multi-modal feature fusion for object detection using neighbourhood component analysis and bounding box regression

Authors: Anamika Dhillon; Gyanendra K. Verma

Addresses: Department of Computer Engineering, NIT Kurukshetra, India ' Department of Computer Engineering, NIT Kurukshetra, India

Abstract: Object detection has gained remarkable interest in the research area of computer vision applications. This paper presents an efficient method to detect multiple objects and it contains two parts: 1) training phase; 2) testing phase. During the training phase, firstly we have exploited two convolutional neural network models namely Inception-ResNet-V2 and MobileNet-V2 for feature extraction and then we fuse the features extracted from these two models by using concatenation operation. To acquire a more compact presentation of features, we have utilised neighbourhood component analysis (NCA). After that, we classify the multiple objects by using SVM classifier. During the testing phase, to detect various objects in an image, a bounding box regression module is proposed by applying LSTM. We have performed our experiments on two datasets: wild animal camera trap and gun. In particular, our method achieves an accuracy rate of 97.80% and 97.0% on wild animal camera trap and gun datasets respectively.

Keywords: deep convolution networks; object detection; neighbourhood component analysis; NCA; support vector machine; SVM; long short-term memory; LSTM.

DOI: 10.1504/IJBIDM.2023.131795

International Journal of Business Intelligence and Data Mining, 2023 Vol.23 No.1, pp.73 - 99

Received: 10 Sep 2021
Accepted: 07 Mar 2022
Published online: 03 Jul 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article