Temporal Coherence Analysis for Intelligent Headlight Control

During nighttime high beams are sparsely used by drivers even when required by the traffic situation. Thus, the intelligent automatic control of vehicles' headlight is of great relevance. Since dazzling other drivers must be avoided, detection of oncoming and preceding vehicles is required. The detection must reach such a long distance that only camera based approaches are reliable. In this case, to detect a vehicle means identifying its head or tail lights. The main challenge is to distinguish these lights from reflections due to infrastructure elements. In order to confront such a challenge we have developed a nighttime vehicle detection system whose core is a novel classifier-based module which can label each detected target as vehicle or non-vehicle. However, in general it is unrealistic to assume a classifier, or a set of them, providing the perfect detection rate and no false alarms. Therefore, we propose to explore the temporal coherence of the targets clas- sification. The usual approach to implement such a coherence analysis requires multi-target tracking. However, tracking is itself a non error-free difficult task. Thus, in this paper we present a different alternative which doesn't require tracking. In particular, we propose a novel confidence accumulation space where the different targets vote according to a confidence value they obtain from the classifier-based module. Targets reaching a predefined threshold in that space are labelled as vehicle and they keep this label according to a hysteresis process. Current results show that this space allows to properly classify clear targets using one single frame, while only a few frames are required for those targets whose type is more difficult to discern.