Task Updates.

24/01/17
Primeesha: For finding the parking rows in image containing vertical lines only, horizontal projection profile is used. It detects white pixels of a row and stores them in an array. Thresholding this array at 30% of maximum value and taking the second derivative gives the starting and ending coordinates of the row. After cropping the particular row using the above obtained coordinates and applying vertical projection profile, it is thresholded at 30% of maximum value again.The first and last array elements satisfying this criteria are taken as the starting and ending point of the required area of that row. For image conatining horizontal lines only, the above procedure has a minor change ,i.e.,first vertical projection profile is used and later we go for horizonal projection profile.

05/01/17
Primeesha: Here, in order to remove the noisy unwanted lines ,image is smoothened and for enhancing the edges it is sharpened. The LOG operator is used to provide the desired result. Using the morphological operations, edges can be detected .Here I have used opening, i.e., erosion followed by dilation. It eliminates all the pixels in image that are too small to contain the structuring element.

03/01/17
Sherry: Here we try to separate the corners  such that the corners falling inside the parking lot form one cluster and the other corners, if any, in the image are separated out. A centroid value is calculated here from which point the distance to all the corners is measured. Since we specify the division to be 2 regions, 2 centroids are calculated in such a way that the corners lying farther away from one centroid but nearer to another are grouped as a cluster with that centroid.The results are updated in the logbook.

25/09/16
Riya & Sherry:  We are working on the algorithm to detect the presence of parking lots in the combined images of Shi-Tomasi and s-plane. We applied some morphological operations on those images so as to remove the smaller objects and overall smoothen the image. The smoothened image example is uploaded in the logbook. We are currently stuck at a point in the algorithm. We will find out the solution as soon as possible.

Neil:  Houghline transform was a possibility for detecting presence of parking lot. I tried it but the results were not as we were expecting. Currently, I am trying out template matching for detection of cars.

01/09/16
Riya & Sherry:  We combined the results of Shi-Tomasi and modified S plane. The results are updated in Logbook.

27/08/16
Riya, Sherry & Neil:  Color information of the image is best viewed in HSV form. So we converted the images to HSV and then split the planes to observe H, S & V plane separately. By observing the planes, S plane was giving the best information. Therefore we thresholded that plane. The result can be seen in the logbook. Also we tried Hough line transform. We have implemented the basic algorithm in that transform. This experiment is updated the logbook.

22/08/16
Riya & Sherry:  We tried to remove green color from the image. First we have to set a range to create the mask for removing the color. So we took a range nearby to color green and experimented with it. We then applied the mask to the image to get the black portions. The result in this experiment as updated in the logbook is not that satisfactory; we will have to improve the range of green color. We will work on that.

15/08/16
Riya & Sherry:  As was suggested, we looked into removing green color from the image. The result we're obtaining is slightly pinkish image as the Blue & Red planes wouldn't change, so it would give the image pinkish color. Also, we're searching for algorithm to capture the area with the max. number of corners which is obtained from the Shi-Tomasi method. There is an algorithm named as camshift or meanshift algorithm which can be used for this purpose. We will study it and try to implement it.

Neil:  I am working on parallel lines detection and sharpening using custom filters.

11/08/16
Riya, Sherry & Neil:  We tried Harris corner detection on images; the results were not that good. So we used another method- Shi-Tomasi corner detection method. Using this on all the images, we were able to obtain max. frequency of corners in the parking lots. This can be used as one of the process to detect the parking lot area. We are now looking for other processes to use in detection of parking lot.

08/08/16
Riya, Sherry & Neil:  We received some suggestions from SPT ma'am. They are: finding patterns of cars by using STFT; trying out Harris Corner Detection to detect if its a parking lot and using projection profiles to find gaps between the rows. We will start working on it.

28/07/16
Riya, Sherry & Neil:  We tried out an example using thresholding and blob detection. Comparatively the results were better than only otsu's thresholding. The results are posted in the logbook section.

26/07/16
Riya, Sherry & Neil:  We had a talk with SPT ma'am about thresholding and blob detection. We are now mainly focusing on thresholding as blob detection needs a properly thresholded image. The solution that ma'am suggested was using a sharpening filter and otsu's thresholding. We tried out laplacian filter along with otsu's thresholding but the results we're getting are not satisfactory.

23/07/16
Riya: To get images of parking lots, I used Bing Maps. I collected 10 images to create a test bench and updated the same in Test Bench section of the wiki. All the images taken are from 50 feet, 20 meters wide.

Sherry: I learned the basics of Blob Detecion. Tried out a sample code from site mentioned in the resources. It is working correctly. But I'm having trouble using the same for different images. The results aren't satisfactory.

Neil: I tried out types of thresholding i.e Adaptive mean and gaussian thresholding on the test bench images. I also tried couple of filters i.e averaging gaussian and median filters. The results are not good enough. We will talk with SPT ma'am to find solutions.

21/07/16
Riya, Sherry & Neil: We had a talk with SPT ma'am and decided to work on creating a test bench, blob detection and different types of thresholding. The main challenges discussed are updated in the 'Main Challenges' section in the wiki.