TY - CHAP
T1 - Detection and Recognition of Vehicle’s Headlights Types for Surveillance Using Deep Neural Networks
AU - Zaheer, Sikandar
AU - Iqbal, Muhammad Javed
AU - Ahmad, Iftikhar
AU - Khan, Suleman
AU - Khan, Rizwan
PY - 2021/10/2
Y1 - 2021/10/2
N2 - Computer vision is been successfully used in the detection and recognition of any object for surveillance in the open and close environment. However, the detection and recognition of any vehicle headlights from image is not an easy task especially in nighttime. Different vehicles have different type of headlights such as Halogen Lamps, Light Emitting Diodes (LEDs), and High-intensity discharge (HID). To identify each vehicle headlights types is very important to control the traffic rules violations. This research work refers to the recent research in intelligent headlight control system (IHC). The main aim to carry out this research work is to mature the automated traffic control systems. This work focus to mature the nighttime traffic control system. It is difficult to monitor each vehicle on road especially in nighttime, so this proposed system is able to monitor each vehicle and identify the vehicles which are violating the traffic laws. Various problems exist in the recognition and detection of headlights, such as erroneous detection of street lights, reflection of water in rain, sign lights and the reflection plate. Some other techniques are also used for this kind of problems; one of them involves the infrared camera to predict the headlight nature. Infrared cameras are not only costly but also need some technical recourse to operate in right manners. This becomes difficult for traffic control authorities to operate them. The proposed system only uses simple images or videos which can be captured through any simple camera by any non technical resource. This work proposes a vehicle headlights type detection and identification method for surveillance using deep learning model such as single shot multibox detector (SSD) mobilenet in real-time video or images data. To detect and predict the nature of headlight of vehicle, we have to create the versatile type of vehicles headlights data and refine it. After preparing and refining the data, we use SSD model, through which we detect and identify the headlight type of vehicle, then we classify those outcomes into three classes (HID, HELOGEN, LED). The more clear data means the more accurate prediction. That’s why in this system we first focus on refining the data and then process the data on SSD model. By adopting this method we results become better, accurate and become faster as compared to other techniques and methods. This method can easily be used for detecting and identifying vehicle headlights from both image and video data.
AB - Computer vision is been successfully used in the detection and recognition of any object for surveillance in the open and close environment. However, the detection and recognition of any vehicle headlights from image is not an easy task especially in nighttime. Different vehicles have different type of headlights such as Halogen Lamps, Light Emitting Diodes (LEDs), and High-intensity discharge (HID). To identify each vehicle headlights types is very important to control the traffic rules violations. This research work refers to the recent research in intelligent headlight control system (IHC). The main aim to carry out this research work is to mature the automated traffic control systems. This work focus to mature the nighttime traffic control system. It is difficult to monitor each vehicle on road especially in nighttime, so this proposed system is able to monitor each vehicle and identify the vehicles which are violating the traffic laws. Various problems exist in the recognition and detection of headlights, such as erroneous detection of street lights, reflection of water in rain, sign lights and the reflection plate. Some other techniques are also used for this kind of problems; one of them involves the infrared camera to predict the headlight nature. Infrared cameras are not only costly but also need some technical recourse to operate in right manners. This becomes difficult for traffic control authorities to operate them. The proposed system only uses simple images or videos which can be captured through any simple camera by any non technical resource. This work proposes a vehicle headlights type detection and identification method for surveillance using deep learning model such as single shot multibox detector (SSD) mobilenet in real-time video or images data. To detect and predict the nature of headlight of vehicle, we have to create the versatile type of vehicles headlights data and refine it. After preparing and refining the data, we use SSD model, through which we detect and identify the headlight type of vehicle, then we classify those outcomes into three classes (HID, HELOGEN, LED). The more clear data means the more accurate prediction. That’s why in this system we first focus on refining the data and then process the data on SSD model. By adopting this method we results become better, accurate and become faster as compared to other techniques and methods. This method can easily be used for detecting and identifying vehicle headlights from both image and video data.
KW - Computer vision
KW - Deep neural network
KW - Headlight type detection
KW - Object detection
KW - Vehicle headlight detection
UR - http://www.scopus.com/inward/record.url?scp=85116817188&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-77939-9_20
DO - 10.1007/978-3-030-77939-9_20
M3 - Chapter
AN - SCOPUS:85116817188
SN - 9783030779382
SN - 9783030779412
T3 - Studies in Computational Intelligence
SP - 689
EP - 707
BT - Deep Learning for Unmanned Systems
A2 - Koubaa, Anis
A2 - Azar, Ahmad Taher
PB - Springer
CY - Cham
ER -