Highway Traffic Surveillance Over UAV Dataset via Blob Detection and Histogram of Gradient

Systems for traffic monitoring often rely on local platforms. Aerial images provide the flexibility to sense the vehicle location and movements with appropriate resolution over a broader area via mobile platforms. This research study presents a method to monitor traffic flow effectively using aerial images. A novel dataset was used to evaluate the research study. In total 100 images were used, and further each image was divided into a burst of three images. The detection is being done on the first image of each burst whereas the other two images were used for tracking. Vehicle detection mainly depends on blob detection techniques with dynamic thresholding methods. To track vehicles, a shape model is generated for each of the detected vehicles. The model is used for template matching to locate all possible positions of the vehicle. The tracking results are improved by selecting only those matches having the direction and magnitude above a threshold. The accuracy of the detection algorithm in terms of correctness and completeness is 86% and 79% respectively. Through geometric computation, 73.4% of vehicles' shape model was correctly estimated. Further, the estimated shape models were used for tracking which tracked 74.9% of vehicles correctly.