Northrop Grumman organized an Image Recognition Challenge at the University of Maryland, College Park. This competition was aimed at using image processing techniques like #ComputerVision or #DeepLearning to identify and locate emergency vehicles like Police Car, Ambulance and Fire Truck in images and videos.
70% of the score included real world testing. The points were awarded for correctly detecting the emergency vehicles, while deducted for false alarms. The rest of the 30% included innovation, presentation and overall technical achievement.
The end goal of this competition was to build a robust mechanism that would accurately detect the emergency vehicles in a scenario having low visibility, obstruction and unexpected behaviors. We used Transfer Learning and the Faster RCNN pre-trained model. We took the benefit of Data Augmentation by using 5200+ images in our dataset. The sheer amount of data and unique approach allowed us to stand out from all the other competitors.
All of these together led us to win the First Prize of $2000 Scholarship Money! We are thankful to Northrop Grumman for giving us this opportunity to explore our limits.
-
Follow the steps mentioned in the 'Setup Instruction.pdf' file for detailed instructions on setting up your system to test (and even train) our model.
-
Next, Download the pretrained model for detecting the emergency vehicles like firetruck, police car and ambulance from
-
Start testing our trained model on any of the test images that can be picked up online. Follow the instructions mentioned in the 'Testing the Trained Model.pdf' file.