Comparative Study of Detection-based Algorithms for Tracking Multiple Cars Alquine Roy F. Taculin MSU - Iligan Institute of Technology Iligan City, Philippines alquineroy.taculin@gmail.com Kardi Teknomo Ateneo de Manila University Quezon City, Philippines kteknomo@ateneo.edu ABSTRACT Vehicle tracking is becoming important in various applications such as traic management, surveillance control, and in road safety. Most tracking methods take advantage of the prediction-correction strategies to locate the moving target objects where image tracking and detection are tightly coupled to achieve superior results. In this paper, a separate and independent tracking method is used to estimate the moving targets’ trajectories based from points pro- vided by the detectors. Two detection algorithms were compared: The Haar Classiier and the Image Diference algorithms. Exclusion method was used to minimize spatial and temporal hijacking. Ex- periments showed that the Image Diference resulted to a superior tracking performance over Haar Classiier. This is attributed to suiciency and consistency of the detected points provided. The tracking method can be further improved by addressing the spatial and temporal Single Car, Multi-Track (SCMT) occurrences which were observed in this study. CCS CONCEPTS • Computing methodologies → Motion processing; Image processing;• Applied computing → Transportation; KEYWORDS vehicle tracking, single car multi-track, image detection, spatial and temporal hijacking, exclusion method ACM Reference Format: Alquine Roy F. Taculin and Kardi Teknomo. 2018. Comparative Study of Detection-based Algorithms for Tracking Multiple Cars. In Proceedings of Philippine Computing Society Congress (PCSC2018). ACM, New York, NY, USA, 6 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn 1 INTRODUCTION Vehicular traic and road accidents are often the primary problems of emerging economies. In their 2015 report, the World Health Organization estimates more than 10,000 deaths related to road accidents in the Philippines [7]. Identifying vehicles ahead and the traic behavior can be an important aspect in monitoring speed, traic surveillance, and road safety. Recently, there has been a signiicant increase in the application of video-based system in facilitating traic dynamics and traic enforcement [16]. Most of the tracking mechanisms anchor their efectiveness through their ability to correctly detect and identify the target objects in the image. According to Shukla and Saini in [15], every vehicle detection includes two stages: Hypothesis Generation which hypothesizes the possible locations of vehicles in the succeeding frames, and Hypothesis Veriication which veriies the estimates and applies correction if necessary. Thus, tracking is tightly coupled with object detection and vice versa. Figure 1: Hypothesis generation and veriication used for image detection and tracking In the process described in Figure 1, tracking of objects is pur- poseful ś the efort is focused on locating the corresponding set of target objects in the succeeding frames. In this paper, our approach decouples the method of trajectory tracking from image detection. We irst allow the detectors to collect the coordinates of the objects of interest (e.g., car) from a video clip. Then trajectories were estimated based on the spatial and temporal positions of those coordinates. Our tracker operates solely on the given coordinates. This approach promotes independence in both detection and tracking modules. In this paper for example, our tracking module does not distinguish the method used for detecting and collecting coordinates. The tracking module assumes no prior knowledge of the objects being tracked, nor allowed to intentionally locate prior objects in the succeeding frames. In the absence of information about the objects (color, shape, size, orientation) being tracked, the video clip used is limited only to cars passing an expressway. The static camera is mounted about 5 meters above the ground, slightly angled downwards. This enables the detectors to generate regularly-spaced coordinates. The road Proceedings of the 18th Philippine Computing Science Congress (PCSC 2018) 185