Autonomous Driving Perception Stack: LiDAR vs Camera-Only

Overview of Technical Issues:

In camera-only autonomous driving perception systems, the camera units insufficiently detect and measure environmental features under variable lighting conditions (night, glare, adverse weather) and cannot directly measure distance accurately, causing degraded perception reliability and reduced safety margins; the goal is to achieve robust all-weather perception with sufficient range and accuracy for safe autonomous navigation.

Solution directions generated for this problem

Problem Direction 1 :

ImproveSensor sensitivity range
VS
ConstraintSystem power consumption

Inspiration 1 : Cross-domain reference

Application Principle: #19 Periodic action
Cross-domain applicability Assess applicability
Apparatuses and methods for improved encoding of images for better handling by displays
Innovative Solution Refine solution

Event-driven adaptive exposure cycling for power-efficient HDR imaging

Trigger multi-exposure only when lighting variance exceeds threshold
How to solve :
  • Deploy ambient light variance monitor sampling at 10Hz
  • activate dual-exposure HDR fusion only when ΔLux >500 lux/s (tunnel entry/exit, sudden glare), otherwise single-exposure mode
  • Implement three-tier exposure strategy: static scenes (ΔLux <50 lux/s) use fixed 10ms exposure at 5W
  • moderate transitions (50-500 lux/s) use adaptive single exposure at 8W
  • extreme transitions (>500 lux/s) trigger 2-frame HDR (1ms + 50ms) at 18W for 2-second bursts
  • Integrate predictive lighting classifier using front-facing photometer array (3×1 layout, 120° FOV): pre-detect upcoming lighting zones 1.5 seconds ahead via spatial luminance gradient analysis, pre-load exposure settings to avoid processing lag
Expected Effect : Average power 12W (20% below baseline); 0.01-100,000 lux coverage; HDR active <15% driving time
Risk Control :
  • photometer calibration drift over temperature
  • false triggering in dappled shade
  • exposure transition lag causing frame drops

Problem Direction 2 :

ImproveDepth measurement precision
VS
ConstraintSystem power consumption

Inspiration 1 : Cross-domain reference

Application Principle: #28 Mechanics substitution
Cross-domain applicability Assess applicability
Method and apparatus for singulating particles in a stream
Innovative Solution Refine solution

Electromagnetic micro-mirror array for optical depth ranging without neural network

Replace neural depth estimation with optical triangulation
How to solve :
  • Integrate electromagnetic voice coil actuator array (64 micro-mirrors, 5×5mm each) in front of existing camera to project structured dot pattern (850nm IR, 200mW laser diode) onto scene for optical depth triangulation, eliminating 30W neural network processing
  • Voice coil actuators (copper coil + neodymium magnet, 12V 50mA per unit) rapidly scan mirror angles ±15° at 120Hz to cover 5-150m range with <5% depth error via geometric calculation (baseline 120mm, focal length 6mm, disparity resolution 0.1 pixel)
  • Closed-loop control using Hall effect position sensors (tolerance ±0.02mm) ensures mirror angle accuracy ±0.1°, maintaining depth precision across temperature -40°C to +85°C
  • total system power 15W baseline + 3.8W actuators + 0.2W laser = 19W versus 45W neural approach.
Expected Effect : Depth error <5% at 5-150m; power 19W vs 45W (-58%); response time <8ms
Risk Control :
  • voice coil actuator positional drift over temperature cycles
  • laser dot pattern occlusion in rain reducing effective range
  • mirror surface contamination degrading reflection accuracy

Problem Direction 3 :

ImproveEnvironmental adaptability
VS
ConstraintDevice complexity

Inspiration 1 : Cross-domain reference

Application Principle: #24 Intermediary
Cross-domain applicability Assess applicability
Telecommunication and multimedia management method and apparatus
Innovative Solution Refine solution

Adaptive message-queue camera exposure scheduler for all-weather perception

Queue-based exposure scheduler for weather adaptation
How to solve :
  • Implement exposure message queue as intermediary layer between weather sensor and camera — queue buffers 5 pre-calculated exposure profiles (clear/rain/fog/snow/night) indexed by ambient light (lux) and precipitation rate (mm/h), camera polls queue every 100ms and applies profile without real-time computation
  • Deploy single low-cost weather sensor module (ambient light + rain detector, <2W, I2C interface) that classifies conditions into 5 states and writes corresponding profile index to queue — no multi-sensor fusion or cross-calibration needed
  • Train 5 lightweight detection models offline (each optimized for one weather profile, 8-bit quantized, <500MB each) stored in flash
  • camera reads queue index and loads matching model in <50ms — decouples weather sensing from perception processing
Expected Effect : Detection range ≥100m in moderate rain/fog; hardware adds only 1 sensor module; calibration reduced to single-sensor characterization; system complexity increase <15%
Risk Control :
  • weather classification accuracy <90% causes wrong profile selection
  • queue polling latency >100ms degrades adaptation speed
  • model switching creates 50ms perception gap

Problem Direction 4 :

ImproveDetection reliability under variable conditions
VS
ConstraintDevice complexity

Inspiration 1 : Cross-domain reference

Application Principle: #11 Beforehand cushioning
Cross-domain applicability Assess applicability
Systems and methods for adaptive monitoring for an environmental anomaly in a shipping container using elements of a wireless node network
Innovative Solution Refine solution

Dual-stream camera with pre-configured exposure redundancy for all-condition detection

Deploy dual-stream camera architecture with pre-configured exposure redundancy
How to solve :
  • Install two identical camera sensors sharing one optical path via beam splitter (50/50 pellicle mirror, 0.2mm thickness) — Stream A fixed at 100ms exposure + ISO 6400 for 0.01–10 lux night scenes, Stream B fixed at 0.5ms exposure + ISO 100 for 1,000–100,000 lux glare conditions
  • Implement region-based stream selection at pixel level — ambient light sensor (TSL2591, 0.1–88,000 lux range) triggers firmware to partition each frame into 16×12 grid zones, selecting Stream A pixels where luminance <50 lux and Stream B pixels where luminance >500 lux, fusing into single output at 30fps with <2ms latency
  • Apply factory pre-calibration protocol — both sensors calibrated simultaneously against ISO 12233 chart under five lighting conditions (0.01, 1, 100, 10,000, 100,000 lux) during manufacturing, storing lookup tables in EEPROM to eliminate runtime cross-calibration complexity
Expected Effect : False negative rate <5% across 0.01–100,000 lux; no HDR processing power; single-pass calibration; detection range maintained ≥120m in night and glare
Risk Control :
  • beam splitter light loss reduces SNR by 3dB per stream
  • pixel-level fusion artifacts at zone boundaries
  • ambient sensor response lag in rapid lighting transitions

Problem Direction 5 :

ImproveSensor sensitivity range
VS
ConstraintMust not deteriorate

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
Transmission apparatus, method of transmitting image data in high dynamic range, reception apparatus, method of receiving image data in high dynamic range, and program
Innovative Solution Refine solution

Predictive exposure pre-configuration system using navigation and ambient light forecasting

Pre-configure camera exposure before lighting transitions using GPS and map data
How to solve :
  • Integrate GPS/map-based lighting prediction module: query road database for upcoming tunnels, bridges, intersections with known glare patterns 5 seconds ahead
  • Pre-adjust exposure settings 1.5 seconds before transition: set 0.5ms exposure + ISO 100 when exiting tunnel into 100,000 lux sunlight, set 50ms exposure + ISO 3200 when entering 0.01 lux tunnel — sensor ready before condition change
  • Deploy forward-facing ambient light sensor array (3 photodiodes at 15°, 0°, -15° angles, 200Hz sampling) to measure upcoming lighting 80m ahead at highway speed, trigger exposure adjustment via lookup table: 0.01-1 lux→50ms/ISO3200, 1-100 lux→10ms/ISO800, 100-10,000 lux→2ms/ISO200, 10,000-100,000 lux→0.5ms/ISO100, transition time 200ms
Expected Effect : Dynamic range 0.01-100,000 lux covered; saturation rate <2%; detection continuity >98% during transitions; power 18W (prediction module +3W)
Risk Control :
  • map database lighting accuracy insufficient in unmapped areas
  • prediction latency causes 0.3s exposure mismatch during rapid transitions
  • photodiode array calibration drift over temperature range
Patsnap Eureka Solution