Multiple Object Detection Based on Clustering and Deep Learning Methods.

Area of Science:

• Computer Vision
• Data Science

Background:

• Multiple object detection is vital in computer vision but is negatively impacted by noise.
• Real-world data, such as underwater sonar images and 3D LiDAR point clouds, often contain significant noise and outliers.

Purpose of the Study:

• To investigate the effectiveness of clustering algorithms in mitigating noise for improved multiple object detection.
• To enhance the performance of deep learning-based object detection systems using noisy sonar and LiDAR data.

Main Methods:

• Deep learning models were employed to process underwater sonar images and 3D LiDAR point cloud data.
• The outputs from these models were then processed using K-Means clustering and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithms.
• These clustering algorithms were utilized to identify and remove outliers, cluster relevant data points, and refine object detection results.

Main Results:

• Both K-Means and DBSCAN demonstrated effectiveness in removing noise and outliers from the processed data.
• The application of clustering algorithms led to a noticeable improvement in the accuracy and reliability of multiple object detection.
• The proposed method showed robust performance across different data types, including sonar images and LiDAR point clouds.

Conclusions:

• Clustering algorithms, specifically K-Means and DBSCAN, offer a viable approach to enhance multiple object detection in noisy environments.
• The findings suggest significant potential for this noise-reduction technique in critical applications like autonomous driving systems and advanced object recognition.