Object Classification in WAMI

Object Detection and Classification on Wide Area Motion Imagery (WAMI) Data

Classification in Wide Area Motion Imagery (WAMI) is a difficult problem because of the low resolution of the objects. Moving objects such as cars and trucks appear as poorly defined contours ranging in size from 2x3 pixels (pedestrian) to 15x10 pixels (trucks). No color information is present in the WAMI data, which would naturally substantially help with the classification. In the project illustrated, a classification of moving objects into two classes is performed. Vehicles are classified as a "car" or a "truck."  The classification process begins by segmenting out moving objects. Based on the segment regions, we utilize features extracted from the intensity values within that region to classify the object appropriately.

The images below illustrate a typical moving object (low resolution) within the scene.

One of the most significant characteristics that distinguish cars from trucks is the significantly different intensity distributions as shown in the images below.

The results of vehicle classification are shown in the figures below. On the left is an unprocessed image from the original dataset. We were able to segment six object within the scene and classify five as cars and one as a truck. On the right are the results of an image that has been processed through the visibility improvement algorithms. Because of the significant enhancement of the data, we able to to segment 27 vehicles into 26 cars and one truck.