Data collection, especially data annotation, is surprisingly time consuming and costly for vehicle classification. This paper presents an algorithm for the semi-automatic annotation of vehicle type that significantly reduces the time needed to annotate a dataset. Vehicles are automatically detected using a background subtraction GMM. The detected vehicles are classified into four main categories: car, van, bus and motorcycle. A vehicle observation vector is constructed from measurement-based features and an intensity-based pyramid HOG (histogram of orientation gradients). K-means clustering is used to initialize the labels of the collected data set. The output scores of a linear SVM classifier are used to identify low confidence samples, which are then manually annotated, significantly reducing the number of samples needing annotation. Experimental results of synthetic and real data set demonstrate the effectiveness and efficiency of our approach. The method is general enough so that it can be used in other classification problems and domains, e.g. pedestrian detection.
Efficient annotation of video for vehicle type classification
2013-10-01
587218 byte
Conference paper
Electronic Resource
English
Efficient Scene Change Detection and Camera Motion Annotation for Video Classification
British Library Online Contents | 1998
|Semi-automatic annotation samples for vehicle type classification in urban environments
IET | 2015
|Semi‐automatic annotation samples for vehicle type classification in urban environments
Wiley | 2015
|Efficiently Scaling up Crowdsourced Video Annotation
British Library Online Contents | 2013
|Human object annotation for surveillance video forensics
British Library Online Contents | 2013
|