CameraComputerExecutive Summary:Keyword List:Introduction:Summary of ProblemShort History of ProblemMotivation and Goals of the ProjectForecast ResultsWhat Reader Will FindProblem Statement:Overview of Design Solution:Key Features:Components and Specific Functions:Step 1: Difference PicturesStep 2: Lowpass FilterStep 3: Centroid DetectionCentroid Detection AlgorithmStep 4: Generate SquareStep 5: Possible Slow-Moving TargetStep 6: Calculate DirectionDiscussion:Test Procedures:Test Results:Determining Threshold vs. Noise PixelsDetermining Threshold vs. Motion PixelsDetermining Centroid Count vs. TimeTarget speed vs. TimeWeak/Critical Areas:Difference Pictures:Lowpass Filter:Centroid Detection:Possible Slow-Moving Target:Conclusion:Impact of This Work:How Will It Be Used:Recommendation for Future Work:References:Bibliography:Documentation to be Delivered:Reference Documents:■ Sun Ultra 1 Data SheetCost Breakdown:Software Requirements:ITEX Coding Standards■ ITEX written in ANSI C with prototypes for each module (located in <modname.h>■ Separate directories for each module, all software functions have the prefix of the module’s name■ Consistent APIs across various modules, division of the CORE, CM, and PA libraries■ Header files:■ C Files:■ Data Types:ITEX Configuration Utilities■ BINCNF:■ ITEXPREP:■ DMAKE:Hardware Requirements:■ Sun Ultra 1MVC 150/40 Hardware Specifications:MVC 150/40 Host PlatformsVME BoardsIMA Advanced Image ManagerCMC Computational Module ControllerAcquisition ModulesAM-CLR ColorProcessing ModulesCM-CLU Convolver/Arithmetic Logic UnitCM-CLU Benchmarks (AOI 512x480)CM-PA Programmable AcceleratorCM-PA Benchmarks (Based on 512x480 image)CM-HF Histogram/Feature ExtractorMemory & I/O ModulesCM-MMP Median/Morp. ProcessorDisplay ModulesDM-PC Pseudocolor Display ModuleSource Code:Tracking Moving Objects in Real-TimeProgramming a Digital Image Processor for Tracking Moving Objects in Real-Time Team 4: Joe Nguyen Chris HorneEE175 WS99 – 04 – 05June 15, 1999Revision 3.0Technical Advisor: Bir BhanuProject Advisor: Barry ToddEXECUTIVE SUMMARY: 1KEYWORD LIST: 1INTRODUCTION: 2Summary of Problem 2Short History of Problem 2Motivation and Goals of the Project 2Forecast Results 3What Reader Will Find 3PROBLEM STATEMENT: 3OVERVIEW OF DESIGN SOLUTION: 4KEY FEATURES: 5COMPONENTS AND SPECIFIC FUNCTIONS: 5Step 1: Difference Pictures 5Step 2: Lowpass Filter 6Step 3: Centroid Detection 7Centroid Detection Algorithm 7Step 4: Generate Square 7Step 5: Possible Slow-Moving Target 8Step 6: Calculate Direction 8DISCUSSION: 9Test Procedures: 9Test Results: 10Determining Threshold vs. Noise Pixels 10Determining Threshold vs. Motion Pixels 11Determining Centroid Count vs. Time 12Target speed vs. Time 12WEAK/CRITICAL AREAS: 12Difference Pictures: 12Lowpass Filter: 12Centroid Detection: 12Possible Slow-Moving Target: 12CONCLUSION: 13Impact of This Work: 13How Will It Be Used: 13RECOMMENDATION FOR FUTURE WORK: 13REFERENCES: 14BIBLIOGRAPHY: 14DOCUMENTATION TO BE DELIVERED: 15REFERENCE DOCUMENTS: 15■ Sun Ultra 1 Data Sheet 15COST BREAKDOWN: 15SOFTWARE REQUIREMENTS: 16ITEX Coding Standards 16■ ITEX written in ANSI C with prototypes for each module (located in <modname.h> 16■ Separate directories for each module, all software functions have the prefix of the module’s name 16■ Consistent APIs across various modules, division of the CORE, CM, and PA libraries16■ Header files: 16■ C Files: 16■ Data Types: 16ITEX Configuration Utilities 16■ BINCNF: 16■ ITEXPREP: 16■ DMAKE: 16HARDWARE REQUIREMENTS: 17■ Sun Ultra 1 17MVC 150/40 HARDWARE SPECIFICATIONS: 18MVC 150/40 Host Platforms 18VME Boards 19IMA Advanced Image Manager 19CMC Computational Module Controller 20Acquisition Modules 21AM-CLR Color 21Processing Modules 22CM-CLU Convolver/Arithmetic Logic Unit 22CM-CLU Benchmarks (AOI 512x480) 22CM-PA Programmable Accelerator 23CM-PA Benchmarks (Based on 512x480 image) 23CM-HF Histogram/Feature Extractor 24Memory & I/O Modules 25CM-MMP Median/Morp. Processor 25Display Modules 26DM-PC Pseudocolor Display Module 26SOURCE CODE: 27The Final ReportThe Solution to a Detailed Problem StatementExecutive Summary:“A moving object is often distinguished better than a motionless one.”R. POLANA AND R. NELSONAs we shall see, the two dimensional visual motion derived from a sequence of time-varying imagery turns out to be a potentially valuable source of information about the 3Dscene being viewed (D. W. Murray and B. F. Buxton, 1990). Understanding visual motionallows us to achieve such tasks as target acquisition and tracking. However, theemphasis on visual motion as a means of quantitative reconstruction of world geometryhas tended to obscure the fact that motion can be used for recognition (R. Polana and R.Nelson, 1997). Our problem is whether motion recognition can be accomplished usingonly low-level features of motion, and if so, how can it be achieved and what methodshould we use?To solve the problem, we have chosen to start with one of the simplest, yet one of themost useful change detection techniques  difference pictures. The most attractive aspectof the difference picture for motion detection is its simplicity. Difference pictures find theareas in a scene which are changing. An area is usually changing due to the movementof an object. Although change detection results based on difference pictures are sensitiveto noise, the areas produced by a difference picture are a good place from which to startsegmentation. In fact, it is possible to segment a scene with very little computation usingdifference pictures. In dynamic scene analysis, this is the lowest level of analysis (R. Jain,R. Kasturi, and B. G. Schunck, 1995).Our report will show that motion recognition can be achieved using low-level features ofmotion without employing abstract models and without prior identification and tracking ofrepresentative feature points.Keyword List:■ Detect■ Difference■ Filter■ Motion■ Moving■ Object■ Pixel■ RecognitionDocument51■ System ■ TrackIntroduction:Summary of ProblemIn the case of machine vision, it might be possible to use motion as a means of recognitiondirectly, rather than indirectly through a geometric reconstruction. Model-basedapproaches have been proposed by earlier researchers to recognize visual motion usuallyemploying higher-level descriptions. The literature to date has not