Intersection Manager (IM) System

• Intersection Manager (IM) system is an integrated cyber-physical framework that fuses multisensor data, radar, LiDAR, and video inputs to optimize intersection safety and traffic flow.
• The system employs edge computing, real-time SPaT estimation, and blind-spot detection to deliver accurate phase control and low latency communication across mixed traffic.
• Using conflict detection and advisory logic, the IM system manages complex interactions among vehicles, cyclists, and pedestrians, thereby enhancing operational safety and efficiency.

An Intersection Manager (IM) system is an infrastructural–cyber–physical framework designed to coordinate and control the motion of vehicles, cyclists, and pedestrians through an intersection with the goal of improving safety, reducing delays, and enabling advanced interaction between road users and infrastructure. IM systems leverage a combination of multisensor perception, communication protocols, online detection and estimation algorithms, spatiotemporal conflict analysis, and logic-based or optimization-driven decision engines. They serve as a core enabling component for both legacy intersections seeking more safety and for future connected/autonomous vehicle (CAV) environments where fine-grained trajectory coordination, predictive safety margins, and infrastructure-to-vehicle (I2V) intelligence are critical (Grembek et al., 2018).

1. Hardware and Software Architecture

The canonical IM system is composed of four primary subsystems: environmental sensing, data fusion and processing, real-time signal phase and occupancy estimation, and low-latency communication interfaces. A typical deployment features the following elements (Grembek et al., 2018):

• Sensors: IP-cameras (30 Hz, for object and red-light detection), 360° scanning RADARs (77 GHz, 100 m), multi-beam LiDARs (e.g., Velodyne VLP-16), and inductive loop detectors embedded upstream of stop bars (used for lane occupancy and actuation input).
• Edge Computing: Discrete nodes for sensor front-end processing (driver software, time sync), object-level detection and fusion, map-referenced localization, and SPaT (signal phase and timing) prediction.
• Map Data: GIS-based storage of intersection topology, lane/guideway geometry, conflict zone polygons, and blind/occlusion zones.
• Communication Modules: Dedicated Short Range Communication (DSRC/IEEE 802.11p), C-V2X (PC5 sidelink), and high-bandwidth Ethernet for WAN integration.
• Software Stack: Linux systems running perception (OpenCV, PCL), tracking (constant-velocity Kalman filter), GIS-server, modular SPaT estimator, Boolean or optimization-based conflict resolver, and J2735-compliant broadcasting at 10 Hz.

This organization allows for deterministic dataflow: raw sensing → detection/tracking/fusion → object localization → SPaT/blind-zone computation → message packaging → broadcast.

2. Signal Phase, SPaT, and Blind-Spot Estimation

The IM system computes and broadcasts the full intersection phase state vector and associated timing variables at high rate. For each time instant $t \ge t_0$ (where $t_0$ is the onset time of the current phase), the full phase vector is:

$$P(t) = [g_1(t), ..., g_8(t); p_2(t), p_4(t), p_6(t), p_8(t)]$$

where $g_i(t) = 1$ iff vehicular movement $\varphi_i$ has green, $p_j(t) = 1$ iff pedestrian phase $P_j$ is WALK. The expected remaining time in the current phase is estimated as

$\tau(t) = \mathbb{E}[D - (t-t_0) \mid D > (t-t_0)]$

where $D$ is the phase duration (random variable), whose empirical statistics are maintained online. Practical implementation uses exponential smoothing for robust estimation:

$\hat D_{k+1} = \alpha D_k + (1-\alpha) \hat D_k$

with $\hat\tau(t) = \hat D - (t-t_0)$. The estimated $\hat\tau(t)$ (with confidence interval) is published in every SPaT message at 10 Hz. Blind-spot occupancy is detected using fused object tracks (LiDAR, radar, vision), with spatial pre-computation of upstream zones $B_i$ for each conflict zone $CZ_i$:

$$B_i = \{ x \mid 0 < d(x, CZ_i) \le v_{max} \hat\tau \}$$

An occupancy indicator $O_i(t)$ is set by a filtered window over confirmed tracks. Empirical accuracy exceeds 98%, with latencies below 50 ms for detection and below 100 ms for message broadcast.

3. Conflict Detection, Resolution Logic, and Advisories

Conflict analysis comprises a core logic engine fusing the estimated phase vector $P(t)$, predicted remaining time $\hat\tau(t)$, and current blind-zone occupancy $\{O_i(t)\}$ for advisory and warning generation. For every road user (vehicle/cyclist/pedestrian), the IM:

• Identifies the discrete set of conflict zones associated with their projected path based on the GIS topology.
• For each conflict zone, applies a ruleset:
  • If the current phase $P(t)$ forbids the conflicting movement, mark "safe".
  • If the user's line-of-sight covers $CZ_i$ and no object is detected, mark "safe".
  • If the conflicting movement is permitted but $O_i(t)=0$, mark "safe".
  • Otherwise, label zone as "unresolved".

Unresolved conflicts with nonzero $\hat\tau$ trigger action advisories—typically, deceleration, "STOP–blind-spot occupied", or phase-hold for right-turn-on-red. For protected-intersection extensions, IM fuses explicit smartphone or tag inputs from cyclists to adapt green-extensions for non-motorized users.

4. I2V Communication Protocols and Message Structure

IMs standardize information exchange via a hybrid of industry and domain-specific formats. The message structure can be represented as:

{ "intersectionID": 2409, "timestamp": "2021-09-01T12:34:56.789Z", "fullPhase": { "φ1":0, "φ2":1, ..., "φ8":0, "P2":0, "P4":1, "P6":0, "P8":0 }, "timeToChange": 7.2, "timeConfidence": 1.5, "blindOccupancy": { "BZ1":0, "BZ2":1, ... } }

Messages are broadcast at 10 Hz via DSRC/IEEE 802.11p or C-V2X, with application-layer encoding compatible with SAE J2735 SPaT standards, extended for intersection collision avoidance (ICA) fields. Update rates, reliability (>99.8% at 500 m), and latency (≤100 ms) are validated in deployment-level tests.

5. Use Cases and Operational Scenarios

Two representative use cases illustrate real-world logic execution:

• Automated Vehicle Negotiating Blind-Spot: An AV approaching a right-turn-on-red consults SPaT and blind-zone occupancy. The IM, finding an unresolved cross-traffic conflict (e.g., detected by sensor fusion but not within vehicle's native field of view), issues a "STOP–blind-spot occupied" advisory. The vehicle system halts or delays the maneuver until the conflict resolves—validated via provided logic pseudocode.
• Pedestrian Conflict Handling: A pedestrian detected in the crosswalk (by vision and/or smartphone app) triggers the IM to send V2X and audible warnings to through-moving vehicles in the parallel phase and extends the "DON'T WALK" infringement period, preventing late phase starts by conflicting movements.

Both use cases leverage Boolean conflict resolution, time-to-arrival estimates, and advisory overlays, driving tangible safety improvements over passive right-of-way or vehicle-side-only strategies.

6. Empirical Performance and Retrospective Assessment

Quantitative evaluation is grounded in real-world failure analysis, notably the March 24, 2017 Tempe collision between an automated Uber Volvo and a Honda CRV:

Metric	IM System
SPaT latency	≤100 ms
Time-to-change error	±0.3 s (95% CI)
Blind-spot detection	98% accuracy (precision 96%, recall 99%)
Message reliability	>99.8% @ 500 m

In detailed simulation of the Tempe scenario, the IM would have issued actionable warnings to both vehicles, increasing headway by >2 s and eliminating collision in 1,000 Monte Carlo runs. By closing the four "information holes" of the event (phase awareness, imminent yellow, cross-traffic detection, and smartphone linkage), the IM demonstrates robust utility as both a software-enhanced "protected intersection" and a systemic countermeasure to critical visibility and timing gaps (Grembek et al., 2018).

7. Integration, Cost, and Deployment Considerations

IM systems are designed as retrofits to existing signalized intersections, with a per-site cost on the order of \$25–50 K (including sensing, edge processing, and wireless modules). The system architecture is modular, leveraging commodity harwdare (sensor suites, Linux/ROS compute) and open V2X communication stacks. All processing is real-time and edge-based, with integration paths to municipal traffic-management WANs via Ethernet or fiber backhaul.

By closing critical information gaps at risk-prone intersections, IMs multiply the effectiveness of Vision Zero strategies and CAV-deployment safety investments. Rollout prioritization is recommended for intersections exhibiting high accident density, complex geometry, or mixed vulnerable road user flow.

---

References:

(Grembek et al., 2018) "Making intersections safer with I2V communication"