Hypothesis based vehicle detection for increased simplicity in multi-sensor ACC

Abstract

Absfruct-Systems for adaptive cruise control (ACC) become increasingly complex in case multiple sensors are used. The search space, detection error and run-time may increase substantially due to combinatory explosion of methods and data. This paper presents a method that simplifies fusion between range and vision devices using corresponding sets of hypotheses. A system is proposed that combines three modules: one uses output o f a 24CHz radar device, one uses single images from a monocular camera system; and one uses the image sequence data o f the same system. The radar detection module uses condensation tracking. The vehicle detection module uses scaled symmetry detection. The three modules are fused by sharing sets of hypotheses for detection of vehicles. Results show 96% error reduction with respect to range sensing only and 63% detection increase due to tracking.

Looking at Vehicles on the Road: A Survey of Vision-Based Vehicle Detection, Tracking, and Behavior Analysis

“…In [169], vehicles were detected using a combination of optical flow, edge information, and symmetry; ranged with radar; and tracked using interacting multiple models with Kalman filtering. In [170], symmetry was used to detect vehicles, with radar ranging. In [171], vehicles were detected using HOG features and SVM classification and ranged using radar.…”

Symmetry-Aided Particle Filter for Vehicle Tracking

“…Experimental results show that the use of symmetry information will strongly increase the tracking performance than the conventional color histogram-based particle filters. In addition, the calculation of symmetry and the combination way of different features are very different from the algorithms in [15] and [1]. More importantly, only visual information is needed in this paper, while [15] and [1] used both visual and radar information.…”

“…However, though the symmetry property is a strong feature of vehicles, it is up to now mainly used in the field of vehicle detection. For vehicle tracking, only [1] and [15] recently tried to used the symmetry features, but their results were based on the combination information of vision and radar.…”

“…These approaches are difficult to deal with the "hijack" problem in multiple-object tracking, which will be illustrated in the following section. Recently, [15] and [1] incorporated the symmetry information into the particle filter framework to track vehicles, which show excellent performance with the aid of radar information.…”

Interactive Road Situation Analysis for Driver Assistance and Safety Warning Systems: Framework and Algorithms

“…In terms of obstacle classification and tracking, the most generally used combination manner consists of a camera and a range sensor [14]- [17]. An approach that simplifies the fusion between range and vision sensors using corresponding sets of hypothesis was proposed [16]. In this system, a radar device and a monocular camera are fused by sharing sets of hypotheses for detection of vehicles.…”