Driver Guidance at Car Washes with DL
Keywords: YOLO, DeepStream, Jetson Nano, TensorRT
At car wash sites, ensuring that cars are correctly positioned on the conveyor belt at the entrance is crucial for a smooth washing process. However, some drivers struggle with this alignment, leading to inefficiencies and potential damage. To address this issue, we developed an advanced system to guide drivers' movements using deep learning technologies.
We implemented a sophisticated guidance system that uses ceiling-mounted cameras to monitor and direct the movement of cars. By detecting the position and orientation of cars and their tires, our system provides real-time feedback to drivers, ensuring proper alignment with the conveyor belt.
Our system significantly improves the efficiency and safety of the car wash process by ensuring cars are correctly positioned on the conveyor belt. This innovation not only enhances the customer experience but also reduces operational disruptions and potential damage to vehicles.
Technical Approach:
• Deep Learning Models: Utilized the YOLO (You Only Look Once) model for real-time object detection, integrated with NVIDIA's DeepStream SDK on Jetson Nano devices for efficient processing.
• Data Collection and Annotation: Due to the unique perspective of ceiling-mounted cameras, tires appeared skewed in the images. There was no existing dataset for these skewed tire images, so we created a custom dataset by collecting and annotating images of cars and tires from this viewpoint.
• Model Training: The YOLO model was trained on this annotated dataset. To enhance accuracy, we employed a hard negative mining approach, focusing on difficult-to-classify examples to improve the model's robustness.
• Latency Optimization: To further improve the system's performance, we integrated TensorRT for optimizing inference latency. By leveraging TensorRT's quantization capabilities, we achieved significant reductions in inference time. This optimization ensures that the guidance system operates in real-time, providing immediate feedback to drivers.
WDVA Information
Certification Number
WDVAARKS23
ARKSOFT INC