Car Parking Space Finder Using Deep Learning: Project Report

This project focuses on developing a car parking space finder using deep learning and artificial intelligence to address the growing issue of parking shortages in urban areas. The project utilizes video input, potentially from a webcam, to analyze parking lot occupancy. The core methodology involves breaking down the problem into smaller, manageable tasks, such as detecting parking spaces and identifying cars within those spaces. The proposed solution employs machine learning techniques, including object detection algorithms like HOG, CNN, and advanced deep learning models like Mask R-CNN, Faster R-CNN, or YOLO, to accurately classify parking spaces as occupied or vacant. The system aims to send notifications when a parking space becomes available. The project addresses potential challenges like shadows, reflections, and nighttime visibility. The methodology includes detecting parking spaces, identifying cars, and determining the occupancy status. The final stage of the project involves alerting drivers about available parking spaces via SMS notifications. This assignment provides a detailed overview of the project, including related works, research problems, questions, methodology, and a timeline for completion.

CAR PARKING SPACE FINDER USING DEEP LEARNING
Abstract
In big cities people faces many problems related to occupancies. One of the major
problem faced nowadays is parking problem. With increase in population vehicles are also
increased day by day which leads to shortage of car parking spaces. Thus, to find parking space
in big cities are sometime frustrating. Thus to overcome this situation it is necessary to build
something which can tell or predict if there is any vacant space for parking in the nearby
location. Also due to car parking there could be heavy traffic jam which can lead to severe
accidents. Thus keeping in mind a system need to be build which will be able to break down the
problem.
Research Area Overview
Using recent widely used technology mainly machine learning, deep learning and
artificial intelligence situation can be solved and can be managed with proper instructions. It is
really very much complicate to build such technology using machine learning and deep learning
for this first it is necessary to break down the problem into a sequence for simple tasks. Then
according to the breakdown and it will be easy to pull different tools from the machine learning
toolbox to solve each of the smaller tasks. By chaining together several small solutions into a
pipeline, we’ll have a system that can do something complicated (Acharya, Yan and
Khoshelham, 2018). The data will be a video which will be provided as an input to the machine
learning model which is generally a stream from a webcam pointed out of the window.
Related Works
Through pipeline each frame need to be passed and it should be one at a time. The initial
stage involves the to capture and detects all the possible parking spaces available in the frame of
the video, thus for these it is necessary to know which portion of the image contains the parking
slot so that it will be easy for the program to identify which parking spaces are occupied and
which are not. Then comes the second which is same procedure as the above here it is necessary
to detect the cars in each frame of the video which will eventually help to track the movement
and space occupied by each car from frame to frame (Amato, et al., 2017). The third step is to

determine which of the parking spaces are currently occupied by cars and which aren’t. This
requires combining the results of the first and second steps. Thus the third step is crucial for any
model or pipeline as if there are any vacant space observed in the frame then only it can be said
to any driver that in that particular zone there is space available for parking. And after which the
last step includes to send notification when a parking space becomes newly available and it will
be only possible if there is any car position change between the frames of the video (Kemker,
Salvaggio and Kanan, 2018).
The projected task is bit hard to implement but possible to accomplish each of these steps
a number of different ways using a variety of technologies (Amato et.al, 2016). There’s no single
right or wrong way to build this pipeline and different approaches will have different advantages
and disadvantages. Thus according to the pipeline neural network model will be implemented
where the data in the form of video will be fetched to classify the vacant parking space and the
cars.
Research Problems and Questions
Few problems which can hamper the performance in transfer learning including the
shadows of the buildings on the parking spaces, strong solar reflection from the vehicles,
vehicles parked outside or in between the designated bays by the drivers and the bias of the
training data used. Also during night time it will be challenging for any model to correctly
classify most of the things which raise a question whether it will be useful during the night time
or not. According to the thinking of the buildup the process will be cost effective or not if not
then due to what the cost increases. According to the system huge storing capacity needed to
store the video which is a challenging task and constant monitoring needed which is also
difficult.
Research Methodology
At first parking spaces need to be detected using the camera thus it is important to scan the
image and get back a list of areas that are valid to park in. The vacant spaces can be identified if
there is any non-moving can then it will be easy to spot the places easily (Camero et al., 2018).
Thus if somehow possible to detect cars and figure out which ones aren’t moving between
frames of video, then it can be taken to consideration that those spaces are parking lots.

 Detection of cars can be identified using various deep learning algorithm which includes
train a HOG (Histogram of Oriented Gradients) object detector and slide it over our
image to find all the cars. Another way is to train a CNN (Convolutional Neural
Network) object detector and slide it over our image until we find all the cars. This one
pretty much easy and can give accurate result but not much efficient and the last one is a
newer deep learning approach like Mask R-CNN, Faster R-CNN or YOLO that combines
the accuracy of CNNs with clever design and efficiency tricks that greatly speed up the
detection process (Mundhenk et al., 2016).
 And the last stage of the project is to send sms alert those who want to park their cars
according if noticed that a parking space has been free for several frames of video. There
are many API used nowadays which lets everyone to send an SMS message from
basically any programming language with a few lines of code. Also there are many other
sms apps available, those can be the second option if API is not chosen.
Timeline

Reference
Acharya, D., Yan, W. and Khoshelham, K., 2018. Real-time image-based parking occupancy
detection using deep learning. In Research@ Locate (pp. 33-40).
Amato, G., Carrara, F., Falchi, F., Gennaro, C. and Vairo, C., 2016, June. Car parking occupancy
detection using smart camera networks and deep learning. In 2016 IEEE Symposium on
Computers and Communication (ISCC) (pp. 1212-1217). IEEE.
Amato, G., Carrara, F., Falchi, F., Gennaro, C., Meghini, C. and Vairo, C., 2017. Deep learning
for decentralized parking lot occupancy detection. Expert Systems with Applications, 72,
pp.327-334.
Camero, A., Toutouh, J., Stolfi, D.H. and Alba, E., 2018, June. Evolutionary deep learning for
car park occupancy prediction in smart cities. In International Conference on Learning and
Intelligent Optimization (pp. 386-401). Springer, Cham.
Kemker, R., Salvaggio, C. and Kanan, C., 2018. Algorithms for semantic segmentation of
multispectral remote sensing imagery using deep learning. ISPRS Journal of Photogrammetry
and Remote Sensing, 145, pp.60-77.
Mundhenk, T.N., Konjevod, G., Sakla, W.A. and Boakye, K., 2016, October. A large contextual
dataset for classification, detection and counting of cars with deep learning. In European
Conference on Computer Vision (pp. 785-800). Springer, Cham.