Highway collision avoidance system based on computer vision is one of the hotspots of current intelligent traffic management system research. How to quickly and accurately detect lanes and vehicles in front of a video image in a changing environment is the most critical issue facing such systems. In the past 20 years, many researchers at home and abroad have conducted a lot of research on this problem, put forward a variety of practical algorithms and successfully developed some visual systems. The algorithms used by these systems can be basically divided into binocular vision-based methods, motion-based methods, shape-based methods and knowledge-based methods. The method based on binocular stereo vision is computationally intensive and requires special hardware support; the motion-based method cannot detect stationary targets and has poor real-time performance; the shape-based method is still a problem to be studied due to the establishment of effective training samples; knowledge-based Method, the efficiency is higher when the number of obstacles is small, but the error rate has increased in complex environments.

Aiming at the shortcomings of conventional algorithms, this paper designs a vehicle rear-end warning system based on monocular vision with high accuracy and good stability. It uses a new edge detection algorithm to identify the road ahead, then uses a combination of shadow detection and tracking to identify the vehicle in front, then judges its threat level according to the distance between the front and rear, and finally provides the driver with a corresponding sound and light alarm signal.

1 Working principle of the system

The system hardware includes MCC-4060 CCD camera, VT-121 video capture card, GPS, PC-104 industrial control computer and display terminal. GPS sends the vehicle speed information to the industrial control computer through the serial port. The CCD camera installed behind the windshield in the car sends the image frame to the industrial control computer through the video capture card. After processing and analysis by the software, the front vehicle obstacle is marked on the display terminal Objects and road markings, at the same time judge the danger level according to the speed, distance, etc., and send out corresponding sound and light alarm signals;

The software part of the system includes road detection, road tracking, vehicle detection, vehicle tracking, ranging, decision-making and alarm modules. When the vehicle speed reaches 60km / h, the system starts to process the image sequence collected in real time. For each frame of image, first detect and track the lane white line in the image, and then detect the vehicle within the area of ​​interest determined by the lane. If there is a suspected obstacle vehicle, vehicle tracking is started, and the tracking information is used to further eliminate false alarms. After achieving stable tracking of obstacle vehicles, the distance between the two vehicles and the relative speed of movement are estimated, the threat level is determined, and the corresponding alarm signal is issued.

2 System key technologies

2.1 Road detection

At present, lane line detection algorithms are mainly suitable for environments with sufficient light. Due to the high contrast between the lane line and the road surface, it is easy to use various conventional edge detection operators to obtain clear lane contour information, then select the appropriate threshold to binarize the image, and finally use the Hough transform to identify the lane line. However, in a complex lighting environment, the image will be disturbed by direct light and multiple reflections of objects to form stray light. The image light intensity cannot reflect the abrupt nature of the lane itself, resulting in the inability to correctly detect the lane.

This system uses a lane detection algorithm that uses the difference in optical density to obtain the difference between the lane marking and the pavement reflectance, and then performs nonlinear edge detection and then Hough transform. This algorithm can effectively solve lane detection under complex lighting conditions, and can also be used for lane detection at night.

In addition, the current lane line tracking research mainly adopts the fixed area method or Kalman filtering method to divide the region of interest according to the result of the lane frame detection in the previous frame to track the lane line in real time. However, the fixed region method relies heavily on the correlation of two frames of images, and the region of interest is large, and the real-time performance is poor; while the Kalman filter method of dividing the region of interest is small, it is easy to produce detection errors, which causes tracking error accumulation and tracking accuracy . Therefore, this system adopts a new method of dividing the region of interest by combining the fixed area method and the KaIman filtering method when tracking the lane line.

Generally speaking, only the area below the intersection of the lane boundary lines (ie vanishing point) and between the two lane lines is regarded as the area of ​​interest. Considering that vehicles driving across lanes still pose a threat to the vehicle, the algorithm separates the two lane lines toward the two. The side is shifted by 40 pixels to expand the area of ​​interest to cover the cross-road vehicle.

2.2 Vehicle inspection

The image contains objects in front of the vehicle with a large field of view, such as roads, trees, guardrails, signs, and other vehicles. It is a difficult task to accurately detect the vehicle in front from it, and the vehicle detection module in this article will automatically change according to the image background. Set parameters to adapt to changing road scenes and lighting conditions.

To achieve rapid detection of vehicles, first of all, it is necessary to perform preliminary detection based on the basic characteristics of the vehicle, extract all possible suspected vehicle areas from the image, and then filter out the suspected areas based on other features.

2.2.1 Initial vehicle inspection

The feature used in the initial detection is the vehicle shadow, that is, an area at the bottom of the target vehicle with a significantly lower gray value than the nearby road surface area. Under general environmental conditions, most vehicles have this distinctive feature.

The process of initial vehicle detection is shown in Figure 1. The shadow of the vehicle has the characteristic of gray level mutation like the lane, so the lane detection algorithm can be called to perform binarization on the original image in FIG. 2 (a) to obtain the edge binary image in FIG. 2 (b). At the same time, the original image is also gray-scale binarized to obtain the gray-scale binarized image in FIG. 2 (c). In order to improve the real-time detection, the average gray level of the road surface area near the vehicle is used as the binary threshold. Since both the edge binarized image and the grayscale binarized image include the lower bottom edge of the vehicle, the OR image of the two images can be used to obtain the vehicle shadow image as shown in FIG. 2 (d).

In the shadow image, search line by line from bottom to top to find the line segment where the continuous shadow point exceeds a certain threshold, and draw a rectangular area as the suspected vehicle area with this line segment as the bottom edge. To ensure that the suspected area contains the entire vehicle, the width of the rectangle is slightly wider than the line segment, and the height is given in proportion to the width. To avoid repeated searches, completely remove the shadows in the suspected areas that have been searched. Since various parts of the same vehicle may be detected as suspicious targets, it is necessary to merge the intersecting suspicious areas. Due to the obstruction of the vehicle in front, multiple targets may be regarded as one target, but it has no effect on the safety of the vehicle.


- Built-in Surge Protector Extension Cord to protect different office devices or deluxe home appliances against electrical surges and spikes.

- Multiple function Power Strip With Surge Protector optional USB ports, various electric outlet etc.

- Master Switch controls power to the outlets on the power bar.

- Multiple Surge Protected Electrical Outlet allowing you to plug in various electrical devices.

- Qualified Surge Protector Extension Cord Power Strip allowing you extend the outlet 6 feet (2 meters) away from the wall or even further.

- Pure PC flame retardant housing Electrical Receptacle with Surge Protection.


Power Strip With Surge Protector

Power Strip With Surge Protector, Surge Protected Electrical Outlet, Electrical Receptacle with Surge Protection, Surge Protector Extension Cord, Power Bar with Surge Protection

ZhongShan JITONGLONG Plastic Hardware Co. Ltd. , https://www.toukoo-electronics.com