Development of Real-time Image Processing Program to Detect Obstacles for Unmanned Tractor.

Abstract

Obstacle detection sensor is an important task for development of autonomous vehicles in recognizing its environment. An examination of various research studies on autonomous vehicles/robots shows that there are only five to six different types of effective obstacle detection sensors. These sensors ranged in price from inexpensive to very expensive. Each of them has their own unique advantages and disadvantages for different applications. If a sensor can be effectively used to create accurate maps for the vehicle environment then applying of such sensor on detection of obstacles in farming environment would be possible. The sensors were consisted of CCD camera, Ultrasonic sensors, Scanning laser, 3D Scanning lasers, and Millimeter wave radar. The purpose of this research was to develop the algorithm of real-time image processing that will be used as obstacles avoidance for unmanned tractor. The equipments in this research were divided into two catagories, including hardware and software platforms. The hardware consisted of Yanmar Tractor EF453T types, Acer Aspire 2930 with Express card slot, CCD camera, cctv lens, express card d/34 to fire wire IEEE 1394A Adapter, fire wire cable, tripleks wood, red pointer laser<5mW, LM 7805, resistor 82 ohm, capacitor 100 μF, luxmeter, and akrilik. While the software consisted of Visual C#, Sharp Develop 3.2 version, and Paint Shop Pro 6 version. The procedure used in present study consisted of the following steps, such as identifying a problem through observation, designing a system for obstacle detection system of unmanned tractor, fabrication of obstacle detection system, capturing the RGB (Red, Green, Blue) values in both static, and dynamic conditions, testing the calibration and validation of optical camera, functional testing, testing the static and real-time performances on unmanned tractor. Image processing techniques such as threshold, erotion, and dilation has achieved successfully to get perfect binary image. The white drop in binary image was then used to detect potential obstacle. Our system is capable to recognize an obstacle in tractor path so that it could give a motion command such as advance, turn left, turn right, or stop. For instance, if the tractor detects the obstacle at less than 7 m then the system would give the command to turn left if the obstacle located in right tractor path. In contrast, it would command to turn right if the obstacle position located in the left tractor path. In case of the obstacle is located at throughout the tractor path, the system would give a command to stop. However, if there is an obstacle in the path tractor, but its range is more than 7 m, the system would give the tractor command to advance. The speed of the vehicle in outdoor application of real-time obstacle avoidance system 0.5 m/s and required 1.3 second for one data processing. This system worked perfectly in 500-200 lux (06.15-08.00 am in normal condition). This system can read the range of obstacle with an accurate up to 81.6% for obstacle in range 2 m, 77.1% in 3 m, 71.5% in 4 m, 69.5% in 5 m, 68% in 6 m, and 48.4% in 7 m. The more obstacle position is located, the more inaccurate system reads the obstacle distance. This might be due to some causes such as the position pointer laser move from fix position. In spite of the sensor position is little move it can make the system fail or inaccurate to read the range value of obstacle. Total accurate for the real-time obstacle avoidance found in current study was 67.4 %. The development of real-time image processing for detection obstacle in outdoor is rather difficult because of its solar intensity. When the solar intensity is high, the system failed to recognize the obstacle and it would be danger for the tractor.