DOI : 10.17577/IJERTV10IS010247
- Open Access
- Authors : Kamal S. Chandwani, Samir Tembhare, Ankit Lambhate, Himanshu Jawale, Rohan Ghagare, Parikshit Bramhankar
- Paper ID : IJERTV10IS010247
- Volume & Issue : Volume 10, Issue 01 (January 2021)
- Published (First Online): 05-02-2021
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License:
This work is licensed under a Creative Commons Attribution 4.0 International License
Cars and Pedestrian Detection
1Prof.Kamal S. Chandwani, 2Samir Tembhare, 3Ankit Lambhate, 4Himanshu Jawale,
5Rohan Ghagare, 6Parikshit Bramhankar
1Assistant Professor, Computer Technology
23456Students, K.D.K. College of Engineering Nagpur, Maharashtra, India
Abstract- Cars and Pedestrian detection are widely applied to intelligent video surveillance, intelligent transportation, automotive autonomous driving or driving-assistance systems. We select OpenCV as the development tool for implementation of cars and pedestrian detection in a video segment. This application will be developed in Python using OpenCV.
KeyWords: OpenCV, Machine Learning, Python, Pedestrian Detection, Vehicle Detection
1.INTRODUCTION –
Recently, there is an advance of miniaturization and lower the cost of cameras have preferred the implementation of large-scale networks of the camera. This increasing number of cameras could permit novel signal processing applications which employ multiple sensors in extensive areas. Object tracking is the novel procedure for discovering moving objects beyond time by utilizing the camera in video sequences. Their main aim is to relate the target objects as well as the shape or features, location of the objects in successive video sequences. Subsequently, the object classification and detection are essential for object tracking in computer vision application. Additionally, the tracking is the first step towards locating or detects the moving object in the frame. Followed by this, detected object could be divided as swaying tree, birds, human, and vehicles and so on. Though, in image processing approach object tracking using video sequences, is a challenging task. Furthermore, several issues appear ascribed to occlusion of the object to scene, object to object, complex object motion, real-time processing requirements as well as the improper shape of the object.
However, this tracking has a large number of benefits, few of them are traffic monitoring, robot vision, surveillance and security and video communication, public areas like underground stations, airports, mass events and animation. Thus, the particular application needs optimal trade-off among computing, communication, and accuracy over the network. The revenue related to computing and communication relies on the amount and type of cooperation executed among cameras for data collection, dispensing and processing to confirm decisions and to reduce the estimation errors and ambivalence. Subsequently, this tracking can be explained as the procedure of determining the orientation of object across the time as the object moves throughout a scene.
This is posting importance in the arena of computer vision because of expansion of high- powered computers and the growing need for automated surveillance systems, and it is broadly applied for applications namely automated surveillance, robotics monitoring, human-machine interface, motion-based recognition, vehicle navigation, traffic monitoring and video indexing. A substantial number of such applications require reliable tracking methods which meet real-time restrictions and are challenging and complex with respect to changes of object movement, scale and appearance, illumination of scene and occlusion. The results of tracking could be impacted by the disparity of one among the parameters. Due to tackle the above- explained issues and others in object tracking numerous approaches have been proposed. In this object tracking application, target will be cars and pedestrians. In addition, moving objects tracking is one of the major tasks in computer vision and broadly applied in industrial vision, intelligent transport systems and visual surveillance.
The capability of machines to identify the suspicious object and further identify their activities in a specific environment is an important part of permitting a machine to interact with humans in effective and easy manner. The current approach for analyzing and detecting the suspicious object usually needs exceptional marker connected to the suspicious object that prevents the extensive technology application. In this paper, to study as wll as analyze the previous approach towards object tracking using video sequences through different phases.
Three key steps in video analysis are discussed as follows:
- Identification of targeted object in moving sequence.
- Object tracking based on one frame to another frame.
- Tracking of the object from camera to camera.
- LITERATURE REVIEW-In the previous study most of them have concentrated towards Object detection (Ben Ayed et al., 2015; Najva and Bijoy, 2016; Ramya and Rajeswari, 2016; Risha and Kumar, 2016; Shen et al., 2013; Soundrapandiyan and Mouli, 2015; Viswanath et al., 2015) ,Object tracking (Bagherpour et al., 2012; Foytik et al., 2011; Lee et al., 2012; Poschmann et al., 2014; Yilmaz et al., 2006; Zhang et al., 2016) and Object recognition (Chakravarthy et al., 2015; Gang et al., 2010; Ha and Ko, 2015; Nair et al., 2011) for tracking the object using video sequences. These are discussed as follows. The basic flow diagram of an object tracking shown in figure 1.
Video sequence
Object Object
Recognition
Object
Figure 1 -The Basic flow diagram of Object tracking
- STUDIES RELATED TO OBJECT DETECTIONThe detection of an object in video sequence plays a significant role in many applications. Specifically as video surveillance applications (Amandeep and Goyal, 2015). The different types of object detection are shown in figure 2.
Object
Frame Optical Flow Background
Recursive Non-
Figure 2- Types of object detection method
Table 1 Comparative Study of Object Detection technique
Pixel-wise
Low
High
Easy to implement (Chate et al.,
Subtraction of
2012; Mohan and Resmi, 2014)
Cannot
be
used
for
real-time
ate
applications (Aldhaheri and
Edirisinghe, 2014)
ate
to
high
Cannot cope up with objects as
Object Detection Method Basic Principle Computa tional Time Accur acy Comments Temporal Differencing Current & Sensitive to dynamic changes Background frame (Haritaoglu et al., 2000) Needs background frame with still objects (Mohan and Resmi, 2014) Backgro Frame Current frame is Low to Moder Simplest background Subtraction (Aldhaheri and Edirisinghe, 2014; Haritaoglu et al., 2000)
Und Differen subtracted from Moderate ate to Subtract cing background frame High Ion applications (Mohan and Resmi, 2014) Approxi Simpe subtraction Low to Moder No need for adequate background modeling (Aldhaheri and Edirisinghe, 2014) mate between median moderate ate Median frame & test frame Requires a buffer with recent pixel values (Aldhaheri and Edirisinghe, 2014) Running Based on Gaussian Moderate Moder Much suitable for real-time Gaussian probability density to high Average function of pixels Statistical calculations consumes more time Mixture Based on Moderate Moder Low memory requirement (Zhiqiang et al., 2006) of Multimodal to high Gaussian Distribution well as noise(Tao Zhang et al., 2010) Optical Flow Uses optical flow distribution characteristics of
pixels of object
Moderate to high High This approach offers entire moving data(Krishna et al., 2011) however require more calculations - LITERATURE REVIEWHere the diagram shows foreground segmentation schemes. (a) Original image. (b) Exhaustive scan (just showing 10 percent of the ROIs). (c) Sketch of road scanning after road fitting in euclidean space . (d) Results of v-disparity applied to the same frame.
- WORKFLOW OF SYSTEM-Working flow of system is depicted as below
Fig-Work flow of pedestrian detection system
- PREPROCESSINGThe preprocessing segment includes tasks such as exposure time, gain adjustments, and camera calibration, to mention a few. Moreover the preprocessing consists of tasks for instance exposure time, gain adjustments, and camera calibration etc. Low- level adjustments are normally not illustrate in ADAS literature, some researchers have targeted image enhancement through these systems. There are two approaches monocular [15] and stereo vision based.
We classify the different features as:
General features: Application independent features such as color, texture, and shape. According to the abstraction level, they can be divided into: Pixel-level features: Features calculated at each pixel, for ex. color, location.
Local features: features calculated over the results of subdivision of the image band on image segmentation or edge detection.
Global features: Features calculated over the entire image or just regular sub-area of an image.
Domain-specific features: Application dependent features as fingerprints, human faces and also conceptual features. These features are frequently a synthesis of low-level features for a specific domain.
All features can be classified into low-level features and high-level features. Low-level features are extracted directed from the original images and high- level feature extraction is based on low level features.
- TRACKING-The most developed systems use a tracking module to track detected cars and pedestrians over time. This step has several reasons like avoiding false detections over time, predicting future pedestrians positions, thus feeding the foreground segmentation algorithm with pre-candidates, and, at a higher level, making useful inferences about pedestrian behavior (e.g., walking direction).
- CONCLUSIONIn this paper, review on different object detection, tracking, recognition techniques, feature descriptors and segmentation method which is based on the video frame and various tracking technologies. This approach used towards increase the object detection with new ideas. We have identified and discussed the limitation/future scope of various methods. Also, we have noted some methods which give accuracy but have high computational complexity. Specifically, the statistical methods, background subtraction, temporal differencing with the optical flow was discussed. However, these technique needs to concentrate towards handling sudden illumination changes, darker shadows and object occlusions (Susar and Dongare, 2015).
- FUTURE SCOPE
- Design and simulation of complex video sequence and test them using same tracking algorithm. In the potential scenario, occlusion is used for an object with the same color for the moving objects or else using bigger occlusion with longer occlusion time. Increasing the number of the object help to identify the efficiency and functionality of the tracking algorithm.
- Weight parameters are needed to be added for individual intensity levels of each pixel. In an image, if an intensity value isassigned as foreground based on the current frame then it has less probability that foreground also has similar pixel coordinate so that BG weightage for the pixel is set to the minimum than the initial value. Through adding weightage lower than the initial value provides the advantage of removing the old pixel value with least probability rather than the evolved scene.
- Need to focus towards enhancing the variance data of each channel based on the Mahalanobis distance calculation. By this,
can able to adopt a change in the rapid scene through Euclidean distance algorithm.
- REFERENCES
- LITERATURE REVIEW-In the previous study most of them have concentrated towards Object detection (Ben Ayed et al., 2015; Najva and Bijoy, 2016; Ramya and Rajeswari, 2016; Risha and Kumar, 2016; Shen et al., 2013; Soundrapandiyan and Mouli, 2015; Viswanath et al., 2015) ,Object tracking (Bagherpour et al., 2012; Foytik et al., 2011; Lee et al., 2012; Poschmann et al., 2014; Yilmaz et al., 2006; Zhang et al., 2016) and Object recognition (Chakravarthy et al., 2015; Gang et al., 2010; Ha and Ko, 2015; Nair et al., 2011) for tracking the object using video sequences. These are discussed as follows. The basic flow diagram of an object tracking shown in figure 1.