An Investigational Study on Ensemble Learning Approaches to Solve Object Detection Problems in Computer Vision

Abstract

Object Detection is a challenging task in computer vision, which is used to identify all or required objects in the given images or videos. The object detection tasks are widely used in many real-world image classifications and face recognition applications like self-driving cars and autonomous robots etc. Considering the challenges in object detection, this paper proposes to present a study on ensemble learning-based approaches to solving object detection problems. The proposed study uses the YOLO algorithmic model (You Only Look Once) to formulate the ensemble learning model with multiple YOLOV3 variants (YOLOV3-320-weights, YOLOV3-SPP and YOLOV3-Tiny). This ensemble learning model, formulated in this study (named YOLOV3-ensembled) is a combination of these algorithmic models. This study, initially, predicts the objects using the YOLO variants individually. Then the variants are combined to detect the objects. The experimental setup included the evaluation metrics IoU (Intersection over Union) and mAP (mean Average Precision). The comparative performance analysis of the ensemble model with other individual models is presented in this paper. It is observed from the results that the YOLOV3-320-weight model could predict the objects more accurately with good IoU scores and mAP scores.