In this project, the smart mainly has two main functions, and we designed two demo situations to show each of the functions.
Demo 1: The emergency braking system
The first function is to have the users control the car while the system noticing the surroundings to make sure the car will not hit anything that might damage the car. Basically, the small car was to simulating an ordinary car that is equipped with an auto emergency braking system. The car has three different modes with three types of instruction LEDs. The original is the green mode in which mode the user could control the small car with a computer. The red mode is a mode which the raspberry pi on the small car find that it might hit an object and make a decision to stop the car, in this mode, the LEDs will turn red. The orange mode is the mode between the red mode and the green mode. It appears when the car is stopped and the car is moving backwards for 15 centimetres.
Demo 2: The visual identification motivated car
In this demo, the small car does not need to be controlled. Instead, the car is controlled by itself. We have the camera on the car to identify different objects with different colours and different shapes. Having identified the objects, the camera could return the object name, position, length and width of the identified object. By using the data, the Pi could generate distance information and do the corresponding action to the object being sensed.
The original design is that the car should be able to do the actions above while being able to automatically find its way to bypass a given barrier. However, due to the limited time, this function has not been developed well.
Problems remaining
This system still has several problems that have not been solved.
The first problem is that the camera on the car is not clear and could easily make mistakes if the light source changes. In the tests before the demo, the light from the window could easily break its white balance and make the car blind. In some cases, the car was confused and recognize the signs wrongly.
The second problem is that the sonar distance sensor returns wrong value randomly, making the Pi easier to make the wrong decisions. Even the research group have tried to avoid this kind of errors from the software level, the bug sometime occurs.
Also, the CCTV of the camera is not prepared. We tried to have the user see the vision from the car's angle and show the results of the identifications. However, it turned out that the bandwidth of the Pi cannot hold the video to transfer with wifi, and the video on the computer would be delayed by five to 10 seconds and the video memory would be occupied soon and the program would be killed.
Even though we still have some problems that are not solved, the car could still finish part of the assumed functions during the demo.
The first function is to have the users control the car while the system noticing the surroundings to make sure the car will not hit anything that might damage the car. Basically, the small car was to simulating an ordinary car that is equipped with an auto emergency braking system. The car has three different modes with three types of instruction LEDs. The original is the green mode in which mode the user could control the small car with a computer. The red mode is a mode which the raspberry pi on the small car find that it might hit an object and make a decision to stop the car, in this mode, the LEDs will turn red. The orange mode is the mode between the red mode and the green mode. It appears when the car is stopped and the car is moving backwards for 15 centimetres.
Demo 2: The visual identification motivated car
In this demo, the small car does not need to be controlled. Instead, the car is controlled by itself. We have the camera on the car to identify different objects with different colours and different shapes. Having identified the objects, the camera could return the object name, position, length and width of the identified object. By using the data, the Pi could generate distance information and do the corresponding action to the object being sensed.
The original design is that the car should be able to do the actions above while being able to automatically find its way to bypass a given barrier. However, due to the limited time, this function has not been developed well.
Problems remaining
This system still has several problems that have not been solved.
The first problem is that the camera on the car is not clear and could easily make mistakes if the light source changes. In the tests before the demo, the light from the window could easily break its white balance and make the car blind. In some cases, the car was confused and recognize the signs wrongly.
The second problem is that the sonar distance sensor returns wrong value randomly, making the Pi easier to make the wrong decisions. Even the research group have tried to avoid this kind of errors from the software level, the bug sometime occurs.
Also, the CCTV of the camera is not prepared. We tried to have the user see the vision from the car's angle and show the results of the identifications. However, it turned out that the bandwidth of the Pi cannot hold the video to transfer with wifi, and the video on the computer would be delayed by five to 10 seconds and the video memory would be occupied soon and the program would be killed.
Even though we still have some problems that are not solved, the car could still finish part of the assumed functions during the demo.
Comments
Post a Comment