Abstract— This paper proposes real time implementation on FPGA for moving objects detection and classification using Handel- C language; the results are shown in RGB video format. In the first part of our work, we propose a GUI interface programmed using the Visual C++ that facilitates the implementation for non-initiate users; from this GUI, the user can program/erase the FPGA or change filters or threshold parameters. The second part of this work details the hardware implementation of real-time motion detection algorithm on a FPGA including the capture, processing and display stages using DK IDE. The third part details the algorithm used to classify the moving objects into humans, vehicles and clutter. The targeted circuit was an XC2v1000FPGA embedded on the Agility RC200E board. A PixelStream-based implementation has been successfully performed and completed with a test validation on real-time motion detection and classification. Keywords—FPGA, Handel-C, Motion detection, Object classification, Real Time, video surveillance. I. INTRODUCTION Detection of moving objects in video streams is known to be a significant and difficult research problem [1].There are many methods dealing with this problem, temporal difference [2][3][4][5][6][7], background subtraction [8][9][10][11] and optical flow[12][13] . After successfully detecting the moving objects, the problem of identification and classification of these objects follows automatically. There are two main categories of approaches towards moving object classification: Shape-based identification and Motion-based classification [14]. Showing the results of moving object detection in RGB will greatly improve the performance of object recognition, classification, identification and motion analysis and will help significantly the operators to make decisions. However, the computational complexity and the huge information for video involved in object detection and segmentation makes it difficult to achieve real-time performance on a general purpose K. Sehairi is with the LTSS laboratory, Amar Telidji University, Laghouat, Algeria. An Assistant Professor in Teacher’s Superior School of Laghouat ENS-L, Algeria (phone number: +213-661931582; e-mail: k.sehairi@mail.lagh-univ.dz). C. Benbouchama is an associate professor in Polytechnic Military School, Algiers, Algeria (e-mail: ben_cherrad@yahoo.fr). F. Chouireb is an associate professor with the Electrical Engineering Department, Amar Telidji University, Laghouat, Algeria (e-mail: chouirebfatima@yahoo.fr). CPU. There exist many architectural approaches to this challenge: 1) Application Specific Integrated Circuit (ASIC), 2) Parallel Computing, 3) GPUs, 4) DSPs, 5)FPGAs. Evolving high density FPGA architectures such as those with embedded DSP multipliers, memory blocks and high I/O (input/ output) pin count make FPGAs an ideal solution in video processing applications [3]. In our implementation, the detection of objects in motion is based on image segmentation, carried out by calculating the differential image of two consecutive frames [2][3][7][15]. In order to detect the object in motion, the flow of data acquired from camera will be split in two parallel sub-blocks; the first one will be converted from YCbCr to gray-scale and will be used to feed the analysis block. The second sub-block will be converted from YCbCr to RGB and merged with the result of the analysis sub-block. The video has a resolution of 720x576 and the object in motion will be presented in a bounding box around it. For this implementation, we should respect two main constraints: the real time processing and the resources of the targeted FPGA. In addition to that, we were interested in developing a Graphical User Interface in order to send the bit- file that configures our FPGA, erases it or changes a parameter in filter, like the threshold parameter. This helps non initiate users to use the program without the necessity to know the hardware architecture and the IDE. II. RELATED WORKS Many methods and techniques for motion detection have already been proposed, in [16] they have been classified in three large categories: Background subtraction, temporal difference, optical flow. K.Ratnayake and A.Aishy [3] developed an algorithm for object segmentation and implemented it in Xilinx XC2VP20 using VHDL; they used frame difference algorithm with a spatio-temporal threshold, the design ran at 133Mpixel/s. In another work presented by M.Gorgon, P.Pawlik et al. [8], the authors used the method of Sum of Absolute Differences to detect vehicles for road traffic, the language used was the Handel-C with the PixelStream library of DK Agility, the implementation was done on RC300 board fitted with an FPGA VirtexIIV6000. The results showed that this implementation process 25 standard PAL images in gray scale with resolution of 576x768 in every second, the number of CLBs used is 11%, 5% block RAMs, and 32% of I/O blocks. A Real Time Implementation on FPGA of Moving Objects Detection and Classification K. Sehairi, C. Benbouchama, and F. Chouireb INTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING Volume 9, 2015 ISSN: 1998-4464 160