Visibility Detection Based on the Recognition of the Preceding Vehicle’s Taillight Signals

Abstract

This paper proposes a method for visibility detection based on the recognition of the preceding vehicle's taillight signals using in-vehicle camera and millimeter-wave (mm-W) radar. First, we design two methods of vehicle identification. One is to use Haar-like features and an AdaBoost algorithm to train the vehicle classifier, which is mainly used to identify vehicles without turning on the taillights. The other is to identify vehicles with taillights on by means of taillight segmentation. The two identification methods are combined with a Discriminative Scale Space Tracker (DSST) to track the vehicle in the image acquired by vehicle camera and to measure anthropic visibility with mm-W radar. In addition, we drove a test vehicle on a foggy highway and collected experimental data through in-vehicle camera and mm-W radar. The experimenter observed the movement of the vehicle in front until it disappeared from the field of vision and recorded the distance of the vehicle in front measured by radar at that time as human visibility, which was also used as the ground truth to verify the accuracy of the proposed visibility detection method. The experimental results show that the visibility measured by the proposed algorithm is essentially consistent with the visibility obtained by human eyes, that is, the visibility of vehicles with no taillights, clearance lamp, emergency flasher, or fog lamp tends to rise, with an average accuracy of 88%, 91%, 90%, and 95%, respectively. In contrast to the traditional visibility measurement, this method mainly measures the maximum distance that the driver can observe when the front vehicle is not turned on or different taillights are turned on.

Driving in the Rain: A Survey toward Visibility Estimation through Windshields

“…Vision sensors, such as cameras, also face difculties in recognising objects hidden behind vegetation. In [55], the camera characteristics are combined to calculate the visual distance of the input frame. Te vehicle-mounted camera mimicked the shooting and recording of the front scene in the experiment, while the vehicle-mounted radar measured and recorded the distance, relative speed, and angle of the barriers and vehicles in front of the experimental vehicle.…”

“…Te control component would incorporate trafc categorisation information when training mainly the vehicle break function to optimise it under those trafc conditions. [55] considers heavy rain and uses a visibility range between 150 − 200 m to pose a risk; however, this method can only work for local visibility estimation when the taillights are on and only has a few use cases during the day, for example, if fog is considered. Such algorithms which provide visibility predictors under various weather conditions are used as a parameter in the perception component of the ADAS/ AD functions.…”

Robust visual detection of brake-lights in front for commercialized dashboard camera

“…In most cases, rear-lights detection is performed first for vehicle detection [9,[20][21][22][23][24][25][26][27][28] or brakelights detection [10,11,[13][14][15][16][17][18][19]29]. Among them, in feature-based methods, rear-lights regions are mainly detected by extracting color features from a specific color space (i.e., RGB [9,14,15,17,20,27], YCrCb [10], L*a*b* [22], HSV [21,23,28], and YUV [19]) based on their characteristics of being red.…”

“…Recently, deep learning or machine learning based methods have been proposed for rear-lights region detection. Some of these learning-based methods introduce trained models as part of rear-lights region detection [13,16,18,24,29,30]. They use a learning model in performing vehicle detection to improve the accuracy of rear-lights region detection.…”

“…They use a learning model in performing vehicle detection to improve the accuracy of rear-lights region detection. Adaboost classifier [13,18], cascade classifier [24], or deformable part model (DPM) [29] is trained to detect the rear of vehicles from given images. The rear-lights region is searched in a space confined to the rear of the vehicle, making detection easier.…”