Vehicle Detection using Edge

Thread Starter

fey93

Joined Dec 7, 2014
7
so im building a robot which is used to detect various types of vehicles, from sedan,jeep to motorcycles

and the digital signal / image processing is hard for me because im not so good at matlab

so if anyone here is a expert on dsp what should i start if i want to do this mini project for my 2nd year of engineering

regards
 

jpanhalt

Joined Jan 18, 2008
11,087
You might start by defining criteria to distinguish the various categories. For example, what distinguishes a coupe, sedan, and motorcycle. BTW, in the US, Jeep is a brand, not a specific type of vehicle. You may think identifying a motorcycle is easy, but in the US, it has become a real problem for a various state motor vehicle departments. Just look at some of the new vehicles for personal transportation.

The above response is actually a response to both threads on this subject by the TS.

John
 

Thread Starter

fey93

Joined Dec 7, 2014
7
i guess the length from front to back, body case morphology, its tire to body ratio, and also outer components.
a jeep has a square body and a tire behind it with a very distinguished model to it
i guess that IS a brand, i dont know what to classify that car in
if there are anymore please let me know

well i do think its easy because they are pretty different from one another (ex: a vespa and a trail bike). what do you think is the problem in indentifing them?
 

GopherT

Joined Nov 23, 2012
8,009
i guess the length from front to back, body case morphology, its tire to body ratio, and also outer components.
a jeep has a square body and a tire behind it with a very distinguished model to it
i guess that IS a brand, i dont know what to classify that car in
if there are anymore please let me know

well i do think its easy because they are pretty different from one another (ex: a vespa and a trail bike). what do you think is the problem in indentifing them?
It seems like an odd, almost useless project. What is the end benefit? Who would need it. If "usefulness" is a grading criterium, I would give it a zero.

I you tell me, well, "toll booths want to know the difference between motorcycles and cars", then limit your project to that. I think it will be a lot of work to differentiate a honda Accord coupe from a honda accord sedan - and no benefit. Unless you have goals that I cannot think of.

In any case, if you are in charge of defining the project, define something that you have the resources and knowledge to implement.
 

jpanhalt

Joined Jan 18, 2008
11,087
i guess the length from front to back, body case morphology, its tire to body ratio, and also outer components.
a jeep has a square body and a tire behind it with a very distinguished model to it
i guess that IS a brand, i dont know what to classify that car in
if there are anymore please let me know

well i do think its easy because they are pretty different from one another (ex: a vespa and a trail bike). what do you think is the problem in indentifing them?
Toyota RAV4 (note spare tire):
RAV4 Capture.PNG

Jeep Cherokee (no tire on back):
Jeep Cherokee.PNG

As gopher suggested, cars vs. motorcycles might be a good place to start. You will need to consider newer designs that blur the obvious distinction of 4 wheels vs. 2 wheels:

upload_2014-12-7_9-52-12.png

John
 

Thread Starter

fey93

Joined Dec 7, 2014
7
It seems like an odd, almost useless project. What is the end benefit? Who would need it. If "usefulness" is a grading criterium, I would give it a zero.

I you tell me, well, "toll booths want to know the difference between motorcycles and cars", then limit your project to that. I think it will be a lot of work to differentiate a honda Accord coupe from a honda accord sedan - and no benefit. Unless you have goals that I cannot think of.

In any case, if you are in charge of defining the project, define something that you have the resources and knowledge to implement.
well its like my lecturer asked us 0 experienced students in Signal and System Processing / DSP/DIP
to make something
anything

that was my idea for the mini project.
so if you have a better idea i would really appreciate it. Image or Signal wise
because i dont know how far my knowledge of MATLAB can take me

i guess it can be for statistics and which can be more valuable to sell from an automotive industry and so forth

for resources i only have matlab and 0 experience in it.
 

Thread Starter

fey93

Joined Dec 7, 2014
7
You keep saying it is easy. If it were, why do you need to come here for help?
i mean nowadays people can make face recog codings easily and other image codes
and i was hoping you guys which are masters in it, can say its easy

for me? its damn hard :(
i tried but i dont know where to start
which to filter
what to use
 

Thread Starter

fey93

Joined Dec 7, 2014
7
Toyota RAV4 (note spare tire):
View attachment 76755

Jeep Cherokee (no tire on back):
View attachment 76756

As gopher suggested, cars vs. motorcycles might be a good place to start. You will need to consider newer designs that blur the obvious distinction of 4 wheels vs. 2 wheels:

View attachment 76757

John
i guess so but is that in video terms or image capture
my idea was my robot will have a motion sensor which will pick off a object with atleast - velocity and take its image
and process it to differentiate the differents
 

jpanhalt

Joined Jan 18, 2008
11,087
i guess so but is that in video terms or image capture
my idea was my robot will have a motion sensor which will pick off a object with atleast - velocity and take its image
and process it to differentiate the differents
Here's an option, read the license plate, then link to the registration index in the country and find out what type of vehicle it is.

John
 

GopherT

Joined Nov 23, 2012
8,009
no, sorry for my intense ignorance
i was just making a point that people are so good at it
You are assuming, just because a product is available, it must be easy to implement. Unfortunately for you, the products you look at generally took years to develop over dozens of generations of products and they had teams of people doing it. One major team and many sub-teams for hardware, A major team for software with many sub-teams, a major team for basic premice of the device and research (e.g for facial recognition: what physical property of faces (vehicles) to monitor, how to measure, how to compare, error rate, what other features of a face can be monitored to reduce error rate, ...).
How to design the database of all known faces, names and other connecting information, ...

These projects are not something you do in an afternoon. This is a big deal, a lot of work and you need to get a better appreciation of what you can do, what you cannot do by yourself (or with your small team) in a semester of work.

As for project ideas, drill down your bigger project idea and start small. Identify one trait of passing cars (color, speed, or something with the noise they make (can you find the major harmonics of each, run it through a Forrier transform and find the major (possibly unique) frequencies each model makes - visually identify make/model/year. If these ideas are too complicated, drill down to something even easier.

I hope you have not been postponing your semester design proejct until December.

Good luck.
 

MrChips

Joined Oct 2, 2009
30,712
i mean nowadays people can make face recog codings easily and other image codes
and i was hoping you guys which are masters in it, can say its easy

for me? its damn hard :(
i tried but i dont know where to start
which to filter
what to use
Because something is readily available does not mean that it is easy.
Internet is readily available. So is any search engine such as google. Same goes for voice and image recognition.
Believe me, all of the above are extremely difficult to implement. If you do not know where to begin I suggest you look for another project that is within your capabilities.

Edit: Sorry, I only just read Gopher's comments. There he has a good idea. How about a project to recognize the color of the car?
 

#12

Joined Nov 30, 2010
18,224
I even meet this in my daily life. "You're good with wood, so make me one of these. It's easy for you, so do it for free."
I don't think so!

Just because you saw one, just because you want one, does not make it easy to do.
 

kubeek

Joined Sep 20, 2005
5,794
To be able to recognize a color of a car, you first need to be able to tell that it is a car. The first two pics in this thread might be easy, the third one probably not.
 

WBahn

Joined Mar 31, 2012
29,978
As others have already pointed out, you are asking for something that is damn, damn, damn hard to do. Since you are the one implementing it, think about just some of the things you would have to do first. Even assuming that the vehicle is perfectly in front of you broadside at a defined distance in perfect lighting, how are you going to implement the algorithm to determine how many wheels it has? Well, that means implementing an algorithm to detect a wheel on a vehicle in a picture. Do you know how to do that? I sure as hell don't (though there are probably some folks here that do). But if, as I suspect, you would be about as lost as I would be trying to just implement that one part, how hard do you think it will be to identify how many doors it has? How would you identify a door?

You need to seriously come down to earth and pick something that you believe you have a pretty good idea of how to viably implement the overwhelming majority of the project before you begin. That will hopefully keep the amount of research you have to do to a manageable level so that you can focus most of the time on building up, implementing, and testing your algorithm.

I can't tell if your project has to be image processing or just more general DSP. If DSP, I would recommend something in the audio realm. Perhaps doing something with audio recordings to improve the quality such as reducing tape hiss on old cassette recordings. Or adding effects, such as echo/reverb, to a recording.
 

GopherT

Joined Nov 23, 2012
8,009
I guess a person must do a project or two and realize what kinds of sensors and information feeds exist directly, and which must be calculated or manipulated from other information. xkcd.com has a few examples...





Or, something like this. Maybe an exaggeration but, the message is clear.

 

jpanhalt

Joined Jan 18, 2008
11,087
Which might get the TS to think again about reading license plates. We know it is doable. They generally have a characteristic shape, color, and position.

As a second phase, the TS might consider whether bumper stickers can be devised that will confound the photo plate readers. ;)

John
 
Top