Thursday 20 March 2025
Researchers have developed a new method for suppressing redundant object detection in images, allowing for more accurate and efficient identification of objects. This improvement could have significant implications for applications such as autonomous vehicles, surveillance systems, and medical imaging.
The traditional approach to object detection involves using algorithms that identify potential objects in an image and then eliminate any duplicates through a process called non-maximum suppression (NMS). However, this method can be prone to errors when dealing with crowded or complex scenes, where multiple objects may overlap or appear similar.
The new technique, developed by scientists at the Institute of Integrated Research, uses a quadratic unconstrained binary optimization (QUBO) approach to suppress redundant detections. QUBO is a mathematical framework that allows researchers to optimize complex problems by converting them into quadratic programs.
In this case, the QUBO formulation was used to identify the most likely objects in an image and eliminate any duplicates based on their spatial overlap and confidence scores. The algorithm was tested on several datasets, including the popular COCO (Common Objects in Context) dataset, which contains images of everyday objects in various settings.
The results showed that the new QUBO-based approach outperformed traditional NMS methods in terms of accuracy and efficiency. For example, when detecting pedestrians in crowded scenes, the QUBO algorithm achieved an average precision of 95%, compared to 85% for traditional NMS.
One of the key advantages of the QUBO method is its ability to handle complex scenes with multiple overlapping objects. Unlike traditional NMS, which can become overwhelmed by the sheer number of potential objects in such scenes, the QUBO algorithm is able to quickly identify and eliminate duplicates, resulting in more accurate and efficient detection.
The researchers also tested their approach on a dataset specifically designed for detecting humans in crowds, known as CrowdHuman. The results showed that the QUBO algorithm was able to detect up to 4% more people than traditional NMS methods, while also reducing the number of false positives by 15%.
The potential applications of this technology are vast and varied. For example, autonomous vehicles could use the QUBO algorithm to identify and track objects in complex traffic scenes, improving safety and efficiency. Surveillance systems could benefit from the algorithm’s ability to detect multiple objects in crowded areas, reducing the risk of false alarms.
In medical imaging, the QUBO approach could be used to improve the accuracy of tumor detection and tracking in cancer patients.
Cite this article: “New Algorithm Improves Object Detection Accuracy and Efficiency”, The Science Archive, 2025.
Object Detection, Image Processing, Qubo Algorithm, Redundant Suppression, Autonomous Vehicles, Surveillance Systems, Medical Imaging, Tumor Detection, Quadratic Unconstrained Binary Optimization, Non-Maximum Suppression
