Road Capture: Developing a Real-Time Vehicle & License Plate Detection System | Rakshithbs | Oct, 2024

As cities face the challenges of increasing vehicle numbers, the demand for advanced systems to ensure safety, manage traffic, and support law enforcement is on the rise. Inspired by this, we delved into the development of a comprehensive solution for real-time vehicle detection and license plate recognition. Our mission was to uncover the possibilities of implementing such systems to provide actionable insights for applications like traffic management, parking systems, and law enforcement. Through our work on these real-world applications, we aimed to gain a deep understanding of the technical complexities and practical aspects involved in bringing these solutions to fruition.

Our objective is to deliver an end-to-end solution capable of processing video frames to detect vehicles in each frame, accurately recognize license plates within those frames, annotate and save the processed video with bounding boxes and information overlays, and extract insights such as vehicle counts and a list of unique license plates detected. Leveraging powerful AI cloud services, our goal was to effectively utilize their robustness and scalability for real-world scenarios.

To kick off our project, we set up the development environment, installed necessary libraries, and securely configured API credentials. After loading the input video, we extracted its key properties like frame rate, width, and height. While processing the video, we extracted one frame per second to balance processing time and detection granularity.

For each frame extracted, we utilized the Plate Recognizer API for license plate detection and reading, showcasing green bounding boxes with recognized plate numbers overlaid for visual reference. Concurrently, we employed Amazon Rekognition for vehicle detection on the same frames, filtering relevant types and marking them with blue bounding boxes. Each frame also displayed additional information such as the total number of vehicles and license plates detected. Subsequently, we saved the annotated frames to a new output video file to ensure smooth playback with real-value overlays.

Our developed system holds significant relevance for various practical applications, including enhanced traffic monitoring, policing support, improved parking management, and streamlined toll collection processes. The automated detection of vehicles and license plates empowers authorities to monitor traffic flow, identify congestion areas, and optimize traffic light coordination for smoother urban transit. For law enforcement agencies, this translates to efficient identification of stolen vehicles and better enforcement of traffic regulations. Deploying automated entry and exit systems in parking facilities improves security and operational efficiency, while automated toll collection reduces traffic congestion and enhances overall driving experiences.

Initially, I faced challenges with Google Cloud Vision API’s object detection accuracy, as it exhibited inconsistency and failed to detect all objects accurately. Upon reassessment, I transitioned to AWS Rekognition, which offered slightly improved object and vehicle detection accuracy.

Another obstacle I encountered was the camera angle in the video footage, which was positioned too low, limiting the visibility of vehicle license plates. A slightly higher angle would have enabled better detection of both vehicles and their plates.

In conclusion, our efforts to develop a system for vehicle and license plate detection have been a rewarding journey filled with challenges and valuable insights. Harnessing the capabilities of computer vision technologies such as the Plate Recognizer API and Amazon Rekognition, we successfully created a system that processes real-time video, accurately identifies vehicles, and extracts pertinent license plate information.

Throughout this project, we learned the importance of optimizing detection accuracy, managing data efficiently, and ensuring real-time responsiveness — essential elements for effective surveillance systems. The ability to filter out irrelevant data and focus on unique license plates underscores the practical applications of our system in various scenarios.

As we look back on our journey, we are excited about the potential of our project and the broader implications it carries for the future of automated surveillance. We welcome researchers, developers, and enthusiasts to engage with us, share their insights, and collaborate on advancing this crucial technological field.

Project Link: Code