Event-based Detection and Tracking
Liu Mi, Ping Xiao · 2024
Due to the advantage of low latency, low power consumption, and high dynamic range of the event camera, it is not necessary to consider the blurring phenomenon of traditional cameras when doing target detection and tracking of moving objects. Therefore, the event camera-based multi-target tracking method generally results in better response time and accuracy[17] This report presents an object-tracking framework based on event count images. The framework contains a YOLO-based detector and multi-threaded single-object trackers based on the Kalman filter. In order to get an event count image dataset for YOLO model training and testing, a proposed event simulator is used to convert a traditional RGB image into a simulated event count image. The converted event image dataset will be fed into the YOLO model for training to get the well-trained detection model. The bounding box data extracted from each video frame by the detector serves as the initial state for the Kalman filter, subsequently enhancing the processing speed through the utilization of Python Multiprocessing. The results obtained by this tracker are used as a benchmark to compare with the results obtained by other two different trackers. The first one uses a Particle filter instead of a Kalman filter for tracking. The second one uses a tracker based on the Kalman filter and the Hungarian algorithm for tracking. The results show that the operating speed increase more than three times after adding the Multiprocessing in Python. The fastest method of the three is the Kalman filter with Multiprocessing in Python.