Real-time positioning of vehicles with Object Detection

In this term paper a solution was developed and evaluated to determine the positions of the vehicles in the Mikrowunderland of the HAW Hamburg in real time. The Mikrowunderland is a miniature exhibition in 1:87 scale, which shows different city districts, sights and landscapes in small models that can be used for scientific purposes.

Object Detection is a technique in image processing that automatically detects and marks specific objects in an image or video. In this term paper Object Detection was used to determine the positions of the vehicles in the microwunderland.

For the experiment, a camera was mounted above the micro wonderland to capture a bird's eye view of the vehicles and the images were processed on a Raspberry Pi. The images were scaled down to a resolution of 300x300x3 pixels and processed using a neural network. The neural network was based on a MobileNetV2 SSD and trained using Transfer Learning and custom images. Data augmentation methods were used to augment the training data.

The neural network was trained using Transfer Learning to improve the recognition of vehicles in the micro wonderland. Transfer learning is a technique where an already trained neural network is used as a base and then further trained for a new task. In this case, a MobileNetV2 SSD was used that had already been trained to recognize objects in images. The network was then further trained with its own images from the micro wonderland to improve the recognition of vehicles.

Additional data augmentation methods were used to enhance the training data. Data augmentation is a technique in which new, synthetic training data is created by modifying existing data. Examples of data augmentation methods include rotating, scaling, and distorting images.

The developed system was able to determine the positions of vehicles in the micro wonderland in real time. By using Object Detection, the system was able to reliably detect the vehicles and accurately determine their positions.