A Radar Guided Vision System for Vehicle Validation and Vehicle Motion Characterization

Abstract

This paper describes a radar-guided monocular lead vehicle ahead of the host vehicle) starts to decelerate in vision system that detects, validates, and tracks the preceding order to perform a turn. Normally, in such scenario, drivers vehicle and thus predicts its lane-change intentions. A vision-have the ability to recognize that by the time they reach the based lane tracking process is developed to create a stable turning place, the primary vehicle should have already tuned motion model in order to map the radar targets to image away from their path. Also, drivers have the ability to adapt coordinates and consequently generate the region of interest their predictions based on the changes of host and target (ROI) to search for a potential preceding vehicle. Model-based a. b object classification algorithms are then applied to validate the ve le se It' mc esire if such palesfcanbe existence of a vehicle in this ROI. Once the detected primary added to ACC or FCW to enhance the system performance. target vehicle, which is in the same lane as the host vehicle, isIn this paper we present a system that uses the radar-cue validated, it will be continuously tracked until it leaves the host guided monocular vision process to validate the radar lane. The spatial-temporal tracking history of the primary detected targets and characterize the moving pattern of the target vehicle is used to infer its intention of changing lanes. in-lane primary target. The system aims to enhance the This prediction results from a characterization process for the performance of ACC and FCW. The objective of the target primary target motion vector. The radar-vision integrated validation is to decide if the target is indeed in the host system has been evaluated on real-world data collected using a vehicle lane and whether it represents an obstacle in the host test vehicle equipped with a radar sensor, vision sensor, and a vehicle path. Non-vehicle targets, such as overhead bridges host processor. and road signs are not considered to be the valid targets and thus will be ignored. The characterization task aims to predict if a primary target is leaving the host lane. A vehicle I. INTRODUCTION that is turning or leaving the host lane is considered to be a V ehicle control and driver awareness system such as target of low threat. In validation process, both moving and [Adaptive Cruise Control (ACC) or Forward Collision stationary targets are considered, whereas in characterization Warning (FCW) systems suffer from poor performance task, only primary moving targets are considered. when reacting to drivable stationary vehicles or precedingIn this radar-guided monocular vision system, vehicle (i.e. lead) vehicles that are turning or leaving the host vehicle validation and characterization are performed by using (the vehicle which hosts the ACC or FCW systems) lane. In pattern classification process on video data. The underlying case of ACC, braking as a result of detecting a stationary supporting video processing tasks involve real-time lane vehicle ...

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

“…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. In [15], monocular vision was used to solve structure from motion, with radar providing probabilities for objects and the ground surface.…”

MmWave Radar and Vision Fusion for Object Detection in Autonomous Driving: A Review

“…Then, the corresponding region of the vision image is extracted according to the ROI. Finally, feature extractor and classifier are used to perform object detection on these images [46] [49] [51] [52] [53] [56]- [58] [61] [62]. The latest literatures use neural networks for object detection and classification [50] [55].…”

Vehicle Detection Techniques for Collision Avoidance Systems: A Review

“…Kalman filtering on radar data was employed for tracking and ranging of identified vehicles. Classifier based detection using HOG, Haar and Gabor features, and range finding by radar was successful in [57], [61]. In another study [58], the input image was analyzed for salient locations using a variety of visual features including orientation, intensity, color, and motion.…”