Fault-tolerant crossroad multi-tracking in urban areas

Details

Supervisors

Macq, Benoît

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en mathématiques appliquées, à finalité spécialisée

Abstract

Due to the expansion of the number of smart cities, city-scale multi-camera vehicle tracking is gaining a lot of interest. Several robust systems have already been implemented in an attempt to address this demand. However, heavy occlusion and faulty detections can greatly reduce the performance of these systems. In this thesis, a simple multi-camera vehicle tracking is implemented with an additional Feedback Reconstruction Algorithm, called interchangeably FRA or feedback. It is a new principle that reconstructs missing detections due to faulty cameras at a crossroad. The algorithm is based on detections made by other cameras. It also uses additional information collected at the previous time frame. The reconstruction’s purpose is to avoid losing a car’s track. The system has been tested in several scenarios where detections have been deliberately removed. The evaluation has been performed on two crossroads using a dataset provided by the AI city challenge [15]. Considering a 50 to 60% missing detection rate on a camera, the system can recover up to 40% of the initial HOTA score (without missing detection) of the faulty camera, and 37.5% of the global HOTA score of the intersection. These results, obtained by simulation, demonstrate the performance and effectiveness of the proposed recovering detection algorithm.