국제대학생창작자동차경진대회 - Lidar 와 Vision- 2016. 12. 22 이재열연구원 1
Contents
Unmanned Solution 1. 자동차는어떻게인식할까? 3
1. 자동차는어떻게인식할까? 현재자율주행차에장착된센서 Lidar (LDMRS) Vision Camera (GT1280) Lidar (LMS111) Radar (Delphi) 4
1. 자동차는어떻게인식할까? 5
1. 자동차는어떻게인식할까? https://youtu.be/51fs0t8hcsq 6
1. 자동차는어떻게인식할까? Camera GPS Lidar * 제공되는플랫폼의그림과다름 7
Unmanned Solution 2. 통신이란무엇인가? 8
2. 통신이란무엇인가? But... Data Sensor USER 9
2. 통신이란무엇인가? 센서에서쓰이는통신 시리얼통신 - 케이블연결로통신 - RS232 / 485 컨버터사용 - 데이터를스트림을바꿔서 ( 직렬화, serialization) 한번에한비트씩전송 - clock 라인을포함 2 라인으로데이터전송 CAN 통신 - 1985 년자동차업체인벤츠의요구에의해독일 Bosch 사가개발 - 차량용네트워크를위해고안된시리얼통신네트워크의통신방식 - 전기적 differential 통신을이용하여전기적인노이즈에강함. - 하나의버스라인에최대 110 개까지의노드연결가능 Ethernet 통신 - 고속근거리통신망에사용 - TCP 와 UDP 통신으로나뉘어짐 - TCP 통신 : 신뢰성이보장되는데이터전송서비스를지원하기위한프로토콜 - UDP 통신 : 신뢰성을보장되지않지만, 빠르게전달가능 10
Unmanned Solution 3. What is a Sensor? 11
시스템구성 ADAS functions overview Adaptive Cruise Control Stop and go Autonomous Emergency Braking Blind Spot Detection Overtake checker Lane Departure Warning Lane keeping Pedestrian & Crossing Traffic Detection Intelligent Headlamp Control Traffic Sign Recognition Short Range LiDAR Sensor Short Range RADAR Sensor 24GHz Multi Functional Camera Stereo Camera 12
Sensor 주변환경인식센서의종류및인식범위 Surround Sensing * 자료 : TEXAS Instruments Making cars safer through technology innovation 13
Sensor ADAS 용센서비교 어두운밤이나악천후에서사용가능 장거리탐지 ( 최대 200m) 형태인식불가능 고가의가격 측정각도가넓음 주변을 3차원으로인지가능 환경의영향을많이받음 고가의가격 형태인식이가능 날씨와시간대에구애를받음 RADAR, LiDAR보다정밀도가떨어짐 RADAR, LiDAR에비해저렴한가격 14 * 출처 : ADAS( 지능형운전자보조시스템 ) 의핵심부품 ( 센서 ) 개발전략 / 한국자동차산업연구소
Sensor 글로벌부품업체들의주변환경인식센서제품현황 * NB : Narrow band * UWB : Ultra wide band Camera RADAR Night Vision Potential Market Supplier Mono Stereo 77RADAR 24 NB RADAR 25 UWB RADAR Far Infrared Near Infrared Rear View Surround View Stability Control Autoliv Continental Bosch Delphi Magna Valeo Denso Gentex Hella TRW Takata 15 * 출처 : Autoliv
RADAR RADAR Application AEB(auto emergency braking) 레이더정보를이용하여최근 vision 시스템위주로개발되어지고있는긴급자동제동시스템연구개발중 - 사용데이터 : Long, Lat Distance, Speed &Acceleration, Azimuth, Amplitude Leader Following 레이더정보를이용하여최근 vision 시스템위주로개발되어지고있는선행차량추종시스템연구개발중 - 사용데이터 : Long, Lat Distance, Speed &Acceleration, Azimuth, Amplitude RADAR 를이용해선행차량을검출하여적정거리설정을통해, 주행속도, 조향각도생성을통해선행차량을자동으로추종하는 기술개발중 16
RADAR Manufactures Delphi ESR Field of view Mid-Range (100m) : +/-45 Long-Range (174m) : +/-10 Scanning frequency Mid Range/ Long Range Target 76.5 GHz 60m/174m 64 Accuracy Range/ Range Rate 0.5 m 0.12 m/s 0.5 Angle Update Rate 0.5 < = 50ms Operating voltage / Power consumption 8v-16v, 24v <1 min <12w mass 575g Size 173.7 x 90.2 x 49.2 (mm) Delphi RSDS +/-75 24 GHz 70m - - 50 (Closing) to +10 m/sec (Opening) - < = 50ms 8v-16v, 24v <1 min, 7w 380g - Manufactures Frequency band 24.05..24.25 GHz 76...77 GHz Distance / Accuracy 1...50 m 0.20 m for point targets 0.25...200 m 0.25 m or 1.5 %@>1 m Azimuth angle augmentation -20...+20 up to -75 +75-8.5...+8.5 far field, -28...+28 close-up range Elevation angle augmentation -6 +6 for -6 db points 4.3 at 6 dbm Speed measurement range -146 km/h...+146 km/h -88 km/h...+265 km/h Mains power supply / Power consumption +9.0 V...16 V DC / app. 4.5 W +8.0 V...27 V DC(12VDC) 7 W at 14 V DC / 7 W at 28 V DC Interface 2 x CAN 1, 2 (car, private) - high-speed 500 kbit/s 1 x CAN 1 - high-speed 500 kbit/s Dimensions (W x H x D) 155 * 131.5 * 26 120 * 90 * 46 Manufactures Frequency range MRR Rear LRR3 MRR Front 76 77 GHz Distance Accuracy 0.5 250 m ±0.1 m Relative speed Accuracy -75 +60 m/s ±0.12 m/s Vision range 30 (-6 db) 5 (-6 db) Modulation Max. number of detected objects Vehicle connector FMCW 32 MQS 8 Pin Cycle time (incl. auto diagnosis) 76 77 GHz Up to 160 m - 45 - - - - 76 77 GHz Up to 100 m - 150 - - - - Typically 80 ms Dimensions (H x W x D) 77 x 74 x 58 (mm) 60 x 70 x 30 (mm) 60 x 70 x 30 (mm) Power consumption typically 4 W - 17 -
RADAR RADAR 데이터획득영상 18
2-12 RADAR RADAR 데이터를이용한 Application 영상 * Source : Automotive RADARS, Robots in Search 19
LiDAR (2D) Manufactures Field of view Scanning frequency Angular resolution Operating range Data Refresh Time Interface Operating voltage / Power consumption Enclosure rating Size IBEO LUX2010 FUSION SYSTEMS 2 layers : 110 4 layers : 85 12.5Hz / 25Hz / 50.5H z 0.125 200m - Ethernet, C AN, RS232 9 ~27V (Average 8W) IP 68 164.5 x 93.2 x 88 IBEO LUX HD 85 12.5Hz / 25Hz / 50.5H z 0.125 90m - Ethernet, C AN, RS232 9~ 27V (Average 8W) IP 68 164.5 x 93.2 x 88 ScaLa 145-0.25 150 m 40/80 ms - 7 W - 105 x 60 x 100 mm Manufactures Field of view Scanning frequency Angular resolution Operating range Response time Interface Operating voltage / Power consumption Enclosure rating Size LMS511-10100 190 25 Hz / 35 Hz / 50 Hz / 75 Hz / 100 Hz 0.167 / 0.25 / 0.333 / 0.5 / 0.667 / 1 0 m... 80 m 10 ms RS-232, RS-422 / E thernet / USB / CANbus 24 V DC / 22 W IP 67 160 x 155 x 185 LD-MRS400001 110 12.5 Hz... 50 Hz 0.125 / 0.25 / 0.5 0.5 m... 250 m - RS-232 / Ethernet / CANbus 9 V ~27 V / 8 W IP 69 94 x 165 x 88 20
LiDAR (3D) Manufactures Field of view Scanning frequency Distance Accuracy Channels Angular resolution Points/sec Operating range Interface Operating voltage / Power consumption Enclosure rating Size 360-16 - 300,000 100m - Low - 100 x 65 VLP-16 360 10Hz < 2cm 32 1.33 700,000 100m Ethernet 9~32V / 12W(12V) IP67 85.3 x 85.3 x 144.2 HDL-32E HDL-64E 360 5-15 Hz < 2cm 64 0.09 1.3 million 120m Ethernet 15V (+/-1.5V) IP67 203.2 x 203.2 x 260.2 Manufactures Field of view Horizontal:360, Vertical:20 (+3 / -17 ) Scanning frequency 10-30 Hz Laser Wavelength 905nm Range Accuracy 1.5cm Angular resolution 0.1 Sensors 8 laser /detector pairs, 3-axis accelerometer Measurement Range 300 m at 80% reflectivity 100 m at 10% reflectivity Interface 1 Gbps Ethernet Operating voltage / Power consumption 9-32 VDC / 20W Enclosure rating IP69K Weight 1kg 21 Size 3.5 dia meter x 3 height
LiDAR LiDAR Application LiDAR 를이용한인식 LiDAR & Camera Traversable area Obstacle speed Obstacle position Object recognition Fusion data Preprocessing Clustering Classification Localization Trajectory Class type Disparity & Depth value Point cloud data 22
LiDAR (2D) 2D Laser Scanner 데이터획득영상 IBEO - LUX IBEO - ScaLa * Source : Youtube 23
LiDAR (3D) 3D Laser Scanner 데이터영상 Velodyne QUANERGY * 장소 : 강남역 * Source : Youtube 24
Camera Digital Image Sensor 외부의전기적신호를 CCD를통해그대로전달하여마지막부분에서전압으로변환 빛에의해발생한전자를그대로출력부까지이동. 노이즈가적어선명한화질 미세한표현과섬세한색상구분가능 Chip 사이즈가작아소형, 경량화가능 소비전력이높고고가의가격 주변부품하나로통합하기어려움. 각화소에서전압으로신호를변환해전송 저전력소비 Blooming effect가없음 빠른이미지저장에적합 소형화가가능 노이즈현상발생이잦음 센서칩의사이즈가큼 정밀한표현력이떨어짐 25
Camera Manufactures Distance range Distance range for light spot(recogniti on) Resolution imager horizontal/ vertical Accuracy distance measuring Viewing angle horizontal/ vertical Response time Power Supply / power consumption Interface Dimensions Multi-Function Camera 4 m (focus range) Up to 500/600m 752pixel / 480pixel ±10m for r<150m, ±50m for r>150m 35 / 20 80ms 2.0W typ 1 x CAN 95.1 x 44.8 x maximum 500 96.7 kbit/s Stereo Camera 60 m 20-30cm (20-30m) - - - - - - - Manufactures Suggested lens FOV over VGA Imager Type No. of Pixels Image area (mm²) Pixel size (µm) Video out Frame Rate (fps) Dynamic Range Oper. Temp 38º CMOS 752 480 4.51 2.88 6.0 6.0 10bit digital 60 60db linear, 100db, nonlinear -45 ~ + 85º C 26
Camera Camera Application Camera 를이용한인식 Send Result Camera Machine Vision Take Action processing Analyzing Lane Tracking Obstacle information Take Image 27
Camera Camera Application Camera 를이용한인식 Input Sensor fusion Output Camera Terrain Information - 모양 - 위치 Object Information - 종류 - 크기 - 위치 - 속도 Feature Information Preprocessing SLAM Map-matching Alignment Road Information 주행가능영역 주행불가능영역 도로곡률 도로고도 도로폭 Lane Information 차선의종류 (Solid/Broken) 차선의개수 차선까지의거리 Preprocessing Segmentation Perception Obstacle Information 장애물의종류 장애물의모양 장애물의속도및방향 28
Camera Camera 데이터영상 MOBILEYE AVM (Around View Monitoring) * 장소 : 언맨드솔루션서초사옥주차장 * Source : Youtube 29
Fusion 통합센서 * 출처 : 콘티넨탈, 프로스트 & 설리번 30
LiDAR Camera, LiDAR 를이용한인식 Pedestrian Feature set data Classification Feature extraction Number of pedestrian Dangers pedestrian Feature extraction Crosswalk Road lane Preprocess Feature matching Sign post Set color model Sign Recognition Recognition sign number Geometry analysis Disparity & Depth value Forward vehicle Feature extraction Classification Vehicle position Vehicle distance Lane model estimation Lane position Number of lane
Fusion 통합센서 32
Fusion 통합센서어플리케이션영상 RACam Lane Tracking Forward Looking Radar Collision Avoidance With Vehicles 33 * Source : Youtube
Autonomous Car 센서퓨전데이터 Velodyne-IMU-Encoder-GPS for Vehicle Localization 34 * Source : Youtube
Unmanned Solution THANK YOU www.unmansol.com / +82-2-3217-6771~2 / ums@unmansol.com 35