Our final project consists of two software modules, one for object detection and another for angle calculations. In our PDR and prototype presentations, we have demonstrated our progress in building the object detection model and discussed our approach in designing the angle calculation algorithm.

Our object detection model now is able to detect antennas in an input image and outputs the coordinates. The computer vision team in Verizon were happy to see the result and wanted to discuss further about the relevant project they are working on that may complement our project. In the prototype presentation, we have received valuable feedbacks from the judges.

Based on the feedbacks and our current progress, we have divided our team into two groups, one to work on object detection and another to work on angle calculation. Our object detection team now is focusing on creating an end to end pipeline that can train the model any time new we have new data available. Additionally, we want to the results to be reproducible and develop a robust performance metrics to evaluate accuracy. Our goal is to automate the entire process of feeding the data, training, making predictions, and evaluating performance.

These are some of the results of our model on the test data.

Figure 1: Object detection model performance that detects antennas with front views.

Figure 2: Object detection model performance that detects antennas with front and side views.