R-CNN Based Vehicle Object Detection via Segmentation Capabilities in Road Scenes

Abstract

In the realm of intelligent transportation systems, vehicle detection and classification stand as pivotal tasks. An effective traffic monitoring system should be capable of detecting, counting, and categorizing moving vehicles. Vehicle classification is a crucial task that can offer insights into road users and help make decisions to reduce congestion, for instance. This paper delves into advanced methodologies for detecting and classifying vehicles on roadways, addressing the limitations of traditional techniques that are often computationally intensive and data acquisition-sensitive. We propose a novel approach that leverages state-of-the-art machine learning and deep learning algorithms to enhance accuracy and efficiency. The proposed model comprises five stages. Initially, all images undergo preprocessing to reduce noise and enhance brightness. In the second stage, the foreground elements are extracted using segmentation techniques. The YOLOv8 algorithm is then used to process these segmented images to identify the vehicles inside them. Next, in the feature extraction phase, the detected vehicles are analyzed using Maximally Stable Estimated Features (MSER), Geometric features and Binary Robust Invariant Scalable Keypoints (BRISK) features. We employ the Recurrent Convolutional Neural Network (R-CNN) classifier for classification. Experimental results from two datasets demonstrate superior performance, with the model achieving an accuracy of 0.94% on the BITVehicle dataset and 0.98% on the Vehicle-OpenImage dataset. Additionally, a comparative analysis was conducted, showcasing the model’s performance against the latest techniques in the field.

Keywords

• Segmentation
• BRISK
• object detection
• vehicle recognition
• YOLOv8
• recurrent convolutional neural network (RNN)