A Robust Approach for Road Users Classification Using Motion Cues
Haider Talib · 2015
Video monitoring of traffic is a common practice in major cities.The data generated by video monitoring has practical uses such as traffic analysis for city planning.However, the usefulness of video monitoring of traffic is limited unless there is also a reliable way to automatically classify road users.This thesis presents a framework that is designed to classify road users into vehicles, cyclists and pedestrians by using the motion cues obtained from their tracks.The road user tracks are obtained using a tracker system such as computer vision techniques.As such, this classification technique does not require additional video or image analysis alongside obtained road user tracks.The separate pieces of information are gained from these motion cues are hereafter called Classifiers.There are nineteen classifiers included in this framework.These classifiers include: average and maximum speeds; average and maximum acceleration; average and maximum deceleration; average and maximum direction; average of change in direction; average of cosine of change in direction; average and maximum area; average and maximum length; average and maximum width; peaks in speed; effective frequency; and the effective weighted average frequency.After obtaining the classifiers' values from the tracked objects' tracks, the information from these classifiers will be assessed and integrated using fuzzy membership approach, which in turn requires prior configurations to be available.This will lead to the final classification of the tracked object.The performance of this framework demonstrated very promising results under different measures.An important contribution of this study is the creation of a robust approach that can integrate different motion cues using fuzzy membership framework.The developed approach also uses parametric classifiers, which do not II depend on the geometry or specific traffic operation of the intersection.This is a key advantage because it enables transferability and improves the practicality and usefulness of the approach.