The ability to detect objects in an outdoor landscape is very important in many emerging fields, such as self driving cars, object tracking for traffic monitoring, pedestrian detection in video surveillance systems, etc. This can be a difficult task considering the magnitude of objects found in urban landscapes, poor lighting or weather conditions, and the presence of many small or hidden objects. For her final project, Tyra used Single Shot Detector (SSD) CNN model to create an object detection application that inputs an image of an outdoor urban street scenes and outputs class names and bounded boxes around four classes of objects: vehicles, road, building and people. This model was created by training the Cityscapes dataset on a Tensorflow2 model that was pre-trained on COCO dataset. Further enhancements to this model include text to speech for an auditory output of the detected classes that can be used for both educational purposes and as an aid for visually impaired.
Urban Street Scenes Object Detection
This CNN model explores object detection of urban street objects, taken from Cityscapes dataset.
Updates
Leave feedback in the comments!
Log in or sign up for Devpost to join the conversation.