In this age of technological advancements, automation is a large part of our everyday life. Distance measurement systems utilizing monocular cameras can only measure for a short range, with a 10-15% inaccuracy and limited robustness. The low accuracy ranging system hinders and slows the process of the machine by a significant amount. Other methods such as the LIDAR or RADAR sensors are prone to disturbances. Thus, we propose the use of a stereo vision image sensor using two monocular cameras connected to a system with a wide baseline. The use of a binocular camera can improve the robustness of the image, accuracy of the distance measurement, and the range of measurement.

We generate our concept through a flow chart of what we are going to do to calculate the distance. Our hardware part are restricted by our sponsor, the total price of two CMOS cameras should be less then 500RMB. On the Fig. 2 the stereo system of two HIKVISION DS-U12i cameras with resolution 1920*1080.

In this project we mainly focus on the generation of our software parts.
On the Fig. 3 is a flowchart of our program.

Two cameras are set parallel with 1 m baseline. This setting simulates a car with cameras in both its front lamps.. For the software part, it basically contains three parts, which are stereo matching, object detection and distance calculation, and we choose python as our program language. The stereo matching consists two sub-functions: feature extraction and combination. We first use SIFT to do the feature extraction, then we can combine two images captured from cameras into one. Then we can detect some objects in this image. For this part, we consider the auto-driving case, and mainly focusing on detecting people and cars.
After detecting some persons or cars, we can just use a formula to calculate the distance. A user interface will show the result to customers Fig. 4.

Validation Process:
For the real distance calculation the laser rangefinder was used for long range measurements. For the stereo matching and camera calibration procedure the size of the block and number of them was measured by a ruler.

Validation Results:
According to the algorithm output:
√Error rate <=5% (< 10 m )
√Cost < 500RMB
√Baseline = 1 m
√Resolution 1920 * 1080p
•Processing time 6-8 s
•Range > 10 m
√ means having been verified and · means to be determined.