Real Time Parking Lot Occupancy Detection by Using Advanced Deep Learning Techniques

Abstract

Real Time Parking Lot Occupancy Detection by Using Advanced Deep Learning Techniques ABSTRACT: A camera is a tool used to record visual data on film, in images, or in videos. However, a smart camera can be recognized as a device that uses recorded video to extract information specific to a certain application. It can be difficult to locate a parking spot in a crowded place because you can't be sure if there are any spare spots nearby. Lack of an empty area causes extra stress before doing the primary task and increases fuel consumption. Our tests on both datasets demonstrate that our solution performs better than the top-performing methods and generalizes them. Our suggested architecture performs similarly to the well-known, three orders of magnitude better, parking lot occupancy detection challenge.

Specification

Description:DESCRIPTIONS:
As the number of automobiles on the road has increased, parking has become a more significant issue, especially downtown. Consequently, there is a high need for efficient parking lot management systems that can deal with these issues immediately. Traditional parking management techniques are failing to ensure optimal space use and reduce congestion due to the limited quantity of parking and the constantly increasing number of vehicles. Convolutional neural networks (CNNs), a type of deep learning technique, have drawn interest because to their potential to revolutionize parking management. The accuracy of occupancy detection that these techniques promise is essential for making well-informed decisions about traffic flow optimization, space distribution, and general urban planning. The development of smart camera technologies that can identify parking lot occupancy has gained popularity recently. Without utilizing a central server, our suggested method completes this operation in real-time immediately on smart cameras. This approach, which is based on deep learning techniques, is scalable, effective, and decentralized. It is predicated on a deep Convolutional Neural Network (CNN) that was created especially for smart cameras. With an emphasis on three different problem types-automatic parking space position recognition, individual parking space classification, and vehicle detection and counting-a number of research have suggested deep learning approaches for parking lot management. The requirement for effective parking space management is the driving force behind the creation of a model for parking lot occupancy detection. It is feasible to maximize parking resource use, improve traffic management, and enhance the entire parking experience for users by precisely determining the occupancy state of parking lots. However, a thorough grasp of the particular difficulties presented by this field is necessary for the effective integration of deep learning in parking management. Complexities including changing lighting, different kinds of vehicles, occlusions, and the need for real-time reaction are introduced by parking scenarios. These difficulties necessitate customized systems that can consistently operate in a variety of settings, offering precise occupancy detection while taking into account the changing dynamics of parking lots. Current approaches to parking lot occupancy identification frequently depend on rudimentary machine learning models or traditional computer vision techniques, which are unable to provide high accuracy in intricate parking situations. Some of the issues described above cannot be resolved by these approaches. This study created a parking lot occupancy detection method utilizing MobileNetV3, a deep CNN classification model with significant architectural changes that improved its accuracy and robustness. Two well-known parking lot datasets, PKLot and CNRPark-EXT, were used to train the created model. Frame-by-frame processing of the incoming video stream separates each frame into patches, which are then classified as either an empty parking space or occupied by a car by the updated MobileNetV3 model. Bounding boxes were created around each parking place, and the classification findings were incorporated into frames. The suggested system's quantitative and qualitative results were experimentally contrasted with those of other well-known classification models. According to the evaluation and testing results, the improved MobileNetV3 model surpassed the other classification models in terms of accuracy and speed, achieving high accuracy. The created parking-space categorization model is effective and applicable to real-world situations including cameras, mobile devices, and edge devices with limited resources.
, Claims:CLAIMS:

1. To determine if parking lot spaces are occupied or vacant, an ideal deep learning model was created.
2. The suggested model substitutes a new activation function that requires less computing for the activation function in the shallow portion of the model, which necessitates substantial calculations.
3. A different, more efficient attention mechanism-the convolutional block attention mechanism-replaces the squeeze-and-excitation attention mechanism used in the original MobileNetV3.
4. Furthermore, blueprint separable convolutions are utilized because they have fewer parameters and require less processing than depth-wise separable convolutions due to the concealed cross-kernel correlations.
5. Our research can be extended by being integrated into a decentralized smart camera system

Documents