MAT-CNN-SOPC: Motionless Analysis of Traffic Using Convolutional Neural Networks on System-On-a-Programmable-Chip

Abstract

Intelligent Transportation Systems (ITS) have become an important pillar in modern "smart city" framework which demands intelligent involvement of machines. Traffic load recognition can be categorized as an important and challenging issue for such systems. Recently, Convolutional Neural Network (CNN) models have drawn considerable amount of interest in many areas such as weather classification, human rights violation detection through images, due to its accurate prediction capabilities. This work tackles real-life traffic load recognition problem on System-On-a-Programmable-Chip (SOPC) platform and coin it as MAT-CNN-SOPC, which uses an intelligent retraining mechanism of the CNN with known environments. The proposed methodology is capable of enhancing the efficacy of the approach by 2.44x in comparison to the state-of-art and proven through experimental analysis. We have also introduced a mathematical equation, which is capable of quantifying the suitability of using different CNN models over the other for a particular application based implementation.Index Terms-Convolutional neural network (CNN), traffic analysis, traffic density, transfer learning, system-on-aprogrammable-chip (SOPC).

TMAV: Temporal Motionless Analysis of Video using CNN in MPSoC

“…In vehicle based assess methodologies, either vehicles are first localized on the road with a background subtraction S. Dey . Frames (Images) from the same Light traffic category of UCSD dataset [8] and associated prediction by a trained CNN model [9] method [10], [16], [17] or the vehicles are localized with moving feature keypoints [11], [12], [18]. Whereas, in holistic approach, a macroscopic analysis of traffic flow is understood through global representation of a scene, which is obtained by accounting for spatio-temporal features except tracking using background subtraction and moving feature keypoints [13], [14], [19].…”

“…In recent times, there has also been emergence of several methods capable of monitoring and analyzing traffic using motionless analysis of videos [4], [9], [20], where videos of traffic are broken into frames instead, and the frames are analyzed for further computation or prediction. The main motivation to utilize methodologies consisting of motionless analysis of video is that it is difficult to stream high-frame rate videos gathered by a large network of interconnected cameras due to bandwidth limitation.…”

“…Hence, streaming lowframe rate videos on these camera networks is very common. In many cases, it is challenging to stream more than 2 frames per second due to the limited bandwidth of the network when these cameras stream over a WIFI network [4], [9]. Moreover, to analyze video in real-time without motion features over a WIFI network is difficult due to communication bandwidth constraint and hence, it is better to analyze the image fames on the camera enabled embedded device itself [9] instead of relying on a server system over th WIFI network.…”

“…In many cases, it is challenging to stream more than 2 frames per second due to the limited bandwidth of the network when these cameras stream over a WIFI network [4], [9]. Moreover, to analyze video in real-time without motion features over a WIFI network is difficult due to communication bandwidth constraint and hence, it is better to analyze the image fames on the camera enabled embedded device itself [9] instead of relying on a server system over th WIFI network. Another motivation to devise such approaches in embedded devices is the afford-ability of such devices instead of employing powerful server systems used for analysis purposes [9].…”

“…Moreover, to analyze video in real-time without motion features over a WIFI network is difficult due to communication bandwidth constraint and hence, it is better to analyze the image fames on the camera enabled embedded device itself [9] instead of relying on a server system over th WIFI network. Another motivation to devise such approaches in embedded devices is the afford-ability of such devices instead of employing powerful server systems used for analysis purposes [9]. Therefore, the approach of analyzing videos without motion features on the embedded device is not just beneficial for categorizing traffic load but could be extended to several computer vision based real-world application that requires analysis of low-frame rate videos.…”